Latest News

A team of researchers, led by Nobel laureate Giorgio Parisi and collaborator Francesco Zamponi, has made a significant breakthrough in solving a long-standing mathematical problem. The solution was found using the generative artificial intelligence (AI) tool Claude, developed by Anthropic. This achievement marks a major milestone in understanding the physics of ‘jamming,’ a phenomenon that had been puzzling researchers for over a decade.

The concept of jamming refers to the sudden transition from a fluid system to a rigid-but-disordered one. To illustrate this idea, imagine a pool table covered with billiard balls. As more and more balls are added, the table eventually becomes congested, and each ball is securely held in place by its neighbors. This disordered, completely frozen situation is known as a jammed state.

The study authors had previously mathematically described jamming and offered numerical solutions in a 2014 paper. However, they noticed that two parameters, $a$ and $b$, would mysteriously always add up to 1. The researchers were puzzled by this relationship and struggled to find an explanation for it.

In separate work, physicist Matthieu Wyart took a different approach but arrived at the same relation. This suggested that entirely new physical concepts were needed to link their work with Wyart’s and explain why $a+b=1$. Despite efforts over the past decade, no progress had been made in finding these new concepts or understanding the reason behind this relationship.

Parisi, who won the 2021 Nobel Prize in physics, decided to revisit the problem. He turned to Anthropic’s Claude AI tool, which he believed could offer a fresh perspective on the issue. After successfully reproducing the 2014 numerical result using Claude, Parisi prompted the AI to prove why $a+b=1$. The initial output contained some errors that required revision, but the fundamental idea was correct.

The researchers were surprised when they found that the solution was hidden directly within the equations themselves. They didn’t need any external physical assumptions or deep connections between functions. This achievement highlights the potential of AI in providing instant access to a vast repository of mathematical training and formal skills that may lie outside one’s usual domain.

Zamponi, who collaborated with Parisi on this project, noted that it is possible for a pure mathematician working full-time on such equations might have spotted the solution. However, he emphasized how Claude gave them instant access to these skills, which they couldn’t see themselves. Zamponi also acknowledged that interacting with AI forces him to reconsider his definitions of reasoning, intuition, and creativity.

The team’s experience with Claude has sparked a new approach to problem-solving in physics. They are now applying this collaborative method to another challenging problem involving the ‘random sequential addition of hard hyperspheres.’ While the AI significantly accelerates writing and optimizing code, Zamponi emphasized that human guidance remains indispensable at least in this case.

The solution to the jamming problem has been published in the Journal of Statistical Mechanics: Theory and Experiment. The study’s authors hope that their work will contribute to a deeper understanding of complex systems and inspire further research in the field.

A new trend is emerging in the world of visual media, where artificial intelligence (AI) is being used to generate images and videos that claim to show historical events. However, experts warn that these creations can be misleading and even deceptive. In Italy’s Pompeii, archaeologists have recently used AI to reconstruct an image of a victim from the 79 C.E. volcanic eruption, marking a significant milestone in the use of this technology for historical purposes.

Roland Meyer, a media scholar at the University of Zurich, believes that the increasing availability of AI tools has led to their widespread adoption without proper consultation with experts. According to him, AI generators can recombine images from the past and create synthetic ones that may not exist but could have existed. This raises concerns about the accuracy and authenticity of these generated ‘historical’ images.

AI-generated videos often blur the line between fact and entertainment, making it difficult for viewers to distinguish what is real and what is fabricated. Historian Sebastian Kubon suggests a simple test: ask whether the video could have been made in the time period it claims to depict. If not, then it’s likely an AI creation.

The video supposedly showing Pompeii can be discredited on this basis alone. Furthermore, it displays clear signs of AI generation, such as people moving in a zombie-like manner and the woman’s appearance looking more modern than what is known about ancient Rome. The ease with which AI tools can create these ‘historical’ images has made it simpler for creators to produce misleading content.

Experts warn that AI generators are trained on existing data and will repeat its biases, resulting in images of the past that primarily reproduce typical patterns, cliches, and stereotypes. As Meyer puts it, these images are not a window into the past but rather ‘a mirror of our present.’ This is particularly concerning when authentic footage exists, as viewers may be misled into believing what they see is real.

When examining AI-generated videos or images, experts recommend looking for visual clues and checking the source. In one video, the camera glides in closer to a group of soldiers holding the US flag, showing their faces as they stare into the distance. However, this looks more like a Hollywood film than authentic documentary footage. Additionally, the watermark gives it away: this video was produced by Veo, a generative AI tool.

The video is based on an iconic 1945 photograph of US soldiers on Iwo Jima, but there is no authentic video footage of this specific moment. All experts emphasized the importance of critical source analysis when dealing with these images and videos. Where does the image come from? Who presents it? For whom and why?

Images and videos always present a particular perspective, as Kubon pointed out. They involve a selection of image fragments and audio sources; in essence, they present one perspective on the story. We’re too quick to assume that just because something was recorded, it’s somehow the truth.

The use of AI-generated images has raised concerns about responsibility. In comparison to fictional scenes in documentaries, where a director or production team decide how to portray the past, AI pulls from large datasets without understanding historical context. It produces something that looks plausible at first glance but is not historically accurate because it matches our expectations.

Jürgen Zimmerer, an Africa history professor at the University of Hamburg, warned that visual sources still carry a great deal of authority and have been problematic even before AI. Images have always been highly controversial sources, evident from dictatorial regimes retouching images in the past. The difference today is the scale at which AI can generate altered and completely fabricated scenes.

To spot AI-generated ‘historical’ images, experts recommend comparing them with verified historical sources such as archives, museums, reputable media or academic experts. A video showing alleged CCTV footage of the Chernobyl nuclear disaster in 1986 is also an AI creation. While authentic photographs exist for Iwo Jima, no authentic footage exists from this moment in Chernobyl.

Such creations raise questions about responsibility and accountability when it comes to using data analysis tools like AI generators. Historians emphasize that history is always an approximation, and pretending there is original film footage can make something seem clear that is not actually clear.

A major day is expected in the world of artificial intelligence as Google prepares to launch its next flagship model, Gemini 3.5 Pro. The highly anticipated release has been widely reported to feature significant upgrades, including a rumored doubling of the context window and the introduction of a new ‘Deep Think’ reasoning layer. However, it’s essential to note that Google has not officially confirmed the date or specifications, so all information below is based on reporting rather than an official announcement.

The Gemini 3.5 Pro was initially expected in June but faced delays due to structural issues discovered during pretraining. According to multiple reports, engineers found problems with recursive tool-calling and SVG generation, which led Google DeepMind to restart the project from scratch. This decision resulted in a later release date but potentially a more capable and stable model.

The delay highlights the high stakes involved in AI development. A few weeks’ difference can make all the difference in this competitive field, where every flagship release is closely watched by both industry professionals and market observers.

A larger context window would have a substantial impact on how models process information. If the reported 2 million-token figure holds true, it means that Gemini 3.5 Pro could ingest vast amounts of data – such as entire codebases or long legal contracts – without losing track of details across all of them.

The addition of ‘Deep Think’ reasoning mode is also significant. This feature would allow models to slow down and work through problems step by step, rather than relying on instinctive answers. The result is a system designed for complex tasks that require chaining multiple actions together with minimal human intervention – the kind of capability every major lab is racing to achieve.

Google’s Gemini line has been moving at an impressive pace. With each new release, the company aims to reclaim its position as the leader in AI development. Gemini 2.5 Pro set a high standard with its 1 million-token context window and strong coding and reasoning scores. Now, Gemini 3.5 Pro is poised to take it up another notch.

The launch of Gemini 3.5 Pro will not be happening in isolation. The past few weeks have seen several flagship releases from rival companies – OpenAI’s GPT-5.6 family on July 9 and xAI’s Grok 4.5 opening its doors to the public around the same time. This crowded AI landscape means that Google has a significant opportunity to make an impact with Gemini 3.5 Pro.

The real battleground for these models lies in their ability to support developers and businesses building on top of them. A larger context window and stronger reasoning capabilities open up new possibilities for tools that can read entire projects, plan complex tasks, and execute them reliably – the kind of functionality that will ultimately determine which model reigns supreme.

For everyday users, the upgrades may seem invisible at first glance. However, it’s in these AI assistants’ ability to support developers and businesses where their true value lies. The tools built on top of Gemini 3.5 Pro have the potential to revolutionize how work is done – making them a crucial part of any business looking to leverage AI for its full potential.

The launch date remains uncertain, with multiple outlets reporting July 17 as the expected release day but emphasizing that this has not been officially confirmed by Google. Until then, it’s essential to wait for an official announcement before getting too excited about specific numbers or features.

If reports are accurate, Gemini 3.5 Pro would represent a significant step forward in AI development – one that directly addresses the capabilities of OpenAI and xAI. However, until we see the actual specifications and performance metrics, it’s impossible to make definitive comparisons between these models.

The American Library Association (ALA) has taken a significant step towards helping libraries navigate the complex landscape of artificial intelligence. The ALA Council has adopted ‘Guidance on the Use of Artificial Intelligence in Libraries’, a comprehensive resource designed to support librarians in developing and revising policies for the ethical, human-centered use of AI.

The guidance was developed by the 30-member Artificial Intelligence Policy Working Group, which consisted of members from across the association. Formed at the 2025 Annual Conference in Philadelphia, the working group brought together experts with diverse perspectives to consider the practical, philosophical, and operational implications of AI for libraries and library workers.

According to ALA President Dr. Maria McCauley, ‘Artificial intelligence is shaping the information landscape, and this guidance from ALA reflects both innovation and the values at the heart of librarianship.’ The document aims to provide a practical starting point for approaching AI responsibly, ethically, and with people at its center.

The adopted guidance affirms ALA’s role in supporting libraries of all types as they adapt to emerging technologies while maintaining a humanistic approach to library service. It gives library workers a values-based foundation for creating, revising, and explaining local AI policies as technology, vendor practices, community needs, and library responsibilities evolve.

Recognizing the increasing importance of AI in library operations, the guidance acknowledges that libraries need adaptable policies to guide implementation now and in the future. To support library workers in interpreting this new resource, ALA will offer virtual training sessions to provide an overview of the guidance and discuss key considerations for libraries.

The adopted guidance is intended as a living document that will be updated regularly to reflect changes in technology, law, professional practice, and community needs. The Information Technology Advisory Committee (ITAC) has been tasked with stewarding this guidance, monitoring developments in AI, and proposing future revisions to the Council as needed.

ALA currently offers a range of publications exploring the evaluation, impact, and implementation of AI in libraries from various angles. These resources are available for purchase through the ALA store, while additional materials can be found on the association’s website.

Honda and Acura are integrating Google Gemini, an artificial intelligence assistant, into certain models that already feature Google software. This upgrade will enable drivers to interact with the system in a more conversational manner than its predecessor, allowing for more natural communication on the road.

The integration of Gemini is designed to enhance connectivity and access to information while driving. Drivers can use voice commands to ask for directions, recommendations, or other types of assistance. The new AI assistant will also be able to answer layered questions and retain context to provide more informed suggestions in the future.

Drivers who are signed into their Google account on compatible models will have access to Gemini’s features. This includes the ability to engage in a ‘genuine conversation’ with the system, rather than relying on strict voice commands as was required for the old Google Assistant. The AI assistant can also assist drivers with tasks unrelated to driving or navigation.

According to Honda, this integration marks an essential step towards improving connectivity and access to information while driving. Jessika Laudermilk, assistant vice president of Honda National Auto Sales, notes that ‘connected technology is no longer a convenience for Honda customers—it’s an essential part of the driving experience.’

The upgraded software will be available on nine Honda models and three Acura models. However, not all trims and model years qualify for the update. The included models are listed below: 2024 Acura ZDX in all trims; 2025-current Acura MDX in all trims; 2026-current Acura ADX A-Spec Advance;

Additionally, the following Honda models will receive the upgrade: 2023-current Honda Accord Touring Hybrid; 2025-current Honda Civic Sedan Si and Sport Touring Hybrid; 2025-current Honda Civic Hatchback Sport Touring Hybrid; 2024-current Honda Prologue in all trims; 2026-current Honda Pilot in all trims;

Lastly, the upgrade will also be available on 2026-current Honda Passport in all trims, as well as 2026-current Honda CR-V Sport Touring Hybrid and 2026-current Honda Prelude in all trims. It’s worth noting that not all trim levels or model years are eligible for this update.

The integration of Google Gemini is part of a broader effort by both Honda and Acura to provide drivers with more comprehensive access to information on the road. By leveraging AI assistants, these manufacturers aim to enhance the driving experience and make it easier for drivers to stay connected and productive while behind the wheel.

A local man has been arrested and charged with multiple felonies for allegedly using artificial intelligence to create child sexual abuse material. The Broward Sheriff’s Office detectives made the arrest last week, targeting a 23-year-old Parkland resident named Ray Lequerique. According to investigators, Lequerique used images of children he personally knew to generate AI-generated content on the Grok platform.

The case highlights the growing concern about online sex offenders exploiting new technologies, including artificial intelligence. Law enforcement agencies are struggling to keep pace with these developments and adapt their strategies accordingly. Greg Schiller, CEO of the Child Rescue Coalition, a South Florida nonprofit that helps law enforcement identify online child predators, notes that AI-generated abuse material is not harmless just because it’s synthetic.

Grok notified the National Center for Missing & Exploited Children after users allegedly generated child sexual abuse material on their platform. This notification was made as required by law. However, investigators say Lequerique continued to create and distribute these materials despite this warning. Less than two months later, detectives arrested him.

Experts warn that AI-generated images can be just as damaging as those taken with a camera. Stuart Pardau, a University of Miami law professor who specializes in artificial intelligence, emphasizes that ‘AI is creating a whole new set of use cases and scenarios that further enable this conduct.’ He stresses the importance of understanding how these technologies are being misused.

Florida recently enacted a law addressing AI-generated child sexual abuse material. The legislation aims to hold perpetrators accountable for their actions, regardless of whether they used real or synthetic images. Schiller supports the new law but also emphasizes the need for parents to educate themselves and their children about online safety.

The case against Lequerique is ongoing, with his attorney arguing that his client’s documented lifelong mental health issues related to an underdeveloped brain at birth could impact how the case proceeds.

The world of artificial intelligence has long been obsessed with benchmark scores. Models are pitted against each other in standardized tests, and a few points higher on any given leaderboard can make or break a model’s reputation. However, at this year’s World AI Conference (WAIC) in Shanghai, Chinese AI players are pushing back against the assumption that these scores tell the whole story.

At WAIC 2026, companies like StepFun and MiniMax showcased their models’ capabilities beyond benchmark scores. Yang Minghui from StepFun emphasized that ‘the benchmark race is not the whole point.’ Instead of focusing on leaderboard rankings, his company prioritizes whether its model actually works well in real-world applications â phones, cars, and robots.

StepFun’s agent operating system STEPX Neo was one such example. The company argues that benchmark homogeneity is particularly problematic for companies like theirs that run multiple vertical-specific models. ‘Each of our models excels in different domains,’ Yang added, ‘no single benchmark can capture that diversity.’

This critique comes as the industry confronts a well-documented problem: benchmark contamination and saturation. As evaluation platform Singularity Moments noted, ‘benchmark contamination is a persistent issue’ and ‘differences within a few points [on MMLU] are often meaningless’ given how easily top models can memorize test data.

MiniMax’s Bai Chuanxu made the case from a different angle. The Shanghai-based firm’s flagship model M3 competes efficiently despite having fewer parameters compared to some rivals. ‘Our model supports images, video and audio natively,’ he explained, ‘but benchmark scores often fail to reflect that and can even create misleading impressions.’

Baidu, whose universal agent DuMate was named a WAIC 2026 official ‘treasure of the hall,’ also sees limits in conventional evaluation. Li Jingqiu, product lead for DuMate, put it differently: ‘The best way to judge a model today is to give it a complex, time-consuming task.’

Baidu’s founder Robin Li suggested that the industry should consider daily active agents as a score, just like daily active users in the age of web. This pragmatic turn â AI is not a monolith â unites these Chinese voices: there’s no one-size-fits-all score.

When models are deployed in real products serving millions of users, the metric that matters is whether the model actually delivers. As Yang Minghui from StepFun said, ‘The real test is just using it.’ This shift away from benchmark scores and toward practical evaluation marks a significant change in how Chinese AI players approach their work.

The industry’s reliance on standardized tests has led to concerns about data leakage and overfitting. Industry analyst site 80AJ observed that ‘static benchmarks are rapidly losing value due to these issues,’ forcing a shift toward human-preference arenas like Chatbot Arena.

The United Nations Secretary-General António Guterres has emphasized that technology must serve people, not the other way around. He made this statement during his speech at the opening ceremony of the World Artificial Intelligence Conference (WAIC) in Shanghai.

Guterres thanked President Xi Jinping for inviting him to the conference and expressed gratitude for China’s warm hospitality in the city of Shanghai. The UN Secretary-General also acknowledged that artificial intelligence can be humanity’s greatest opportunity in the 21st century, but warned that it could also become one of its greatest risks if not managed properly.

The theme of the WAIC reminds us that a brighter future depends on partnerships. Guterres expressed his appreciation for China bringing together international leaders and experts to discuss AI governance and development. He noted that the launch by President Xi Jinping in 2023 of the Global AI Governance Initiative was a strong message of Chinese commitment and leadership, and that the World AI Conference is its natural development.

The UN Secretary-General highlighted the importance of partnership in achieving sustainable development goals through the application of artificial intelligence. He mentioned that AI can accelerate medical breakthroughs, transform education, strengthen food systems and agricultural productivity, and spur new industries and create decent jobs.

Guterres emphasized that technology must be shaped by all humanity, not just a handful of countries or companies. Every nation needs a seat at the table when it comes to governing AI. He noted that this is reflected in the Global Digital Compact adopted by the UN General Assembly with China’s full support.

The Compact made it clear that technology shaping the future of humanity must be shaped by all humanity, not just a few countries or companies. Guterres stated that every nation needs a seat at the table when it comes to governing AI and that this is reflected in the Global Digital Compact adopted by the UN General Assembly with China’s full support.

Guterres highlighted three priorities for ensuring that artificial intelligence benefits all countries: capacity, safety, and sustainability. He emphasized that developing countries bring enormous assets to the age of AI, including talent, ideas, and young populations. Every country should be able to shape this technology to its own needs, with its own skills, data, and languages.

Guterres noted that China has an essential role to play in ensuring that artificial intelligence benefits all countries. He mentioned that over 20 countries, including China, have already responded to his invitation and nominated centers for a UN-supported Global Network for Exchange and Cooperation on AI Capacity Building at the service of developing countries.

The Secretary-General also announced plans to put forward recommendations for a Global Fund for AI with the same purpose as the capacity-building initiative. He asked for full support from governments and technology companies for both initiatives, emphasizing that they are crucial for ensuring that artificial intelligence benefits all countries.

Guterres highlighted the importance of safety in developing AI systems. He emphasized that humans must keep control over every life-and-death decision made by AI systems. The Secretary-General also called on major AI companies to disclose their full environmental footprint and power their systems with renewables by 2030.

The UN Secretary-General concluded his speech by emphasizing that the defining question is not whether artificial intelligence will transform the world, but rather how it will be managed. He urged governments and technology companies to work together to ensure that AI reduces inequalities and expands opportunity for all countries.

When it comes to using local Large Language Models (LLMs), many users have high expectations, often mirroring the performance of cloud-based models like ChatGPT. However, these expectations can be unrealistic, especially when running on modest hardware. In this article, we’ll explore why local LLMs struggle with broad questions and what tasks they excel at, helping you set realistic expectations for their capabilities.

The largest models that can practically run on a mini PC are 8B models, which is significantly smaller than the leading cloud-based models like DeepSeek-V3, published in 2024, boasting an impressive 671 billion total parameters. It’s likely that current frontier models are at least as large or even larger. This disparity highlights the limitations of local LLMs when compared to their cloud-based counterparts.

When I first installed a local LLM on my mini PC, I expected it to work similarly to ChatGPT. However, this wasn’t the case. The model took an age to respond to broad questions and even then produced disappointing results. It soon became apparent that using a local LLM required a different approach.

Leading cloud-based models excel at answering complex, open-ended questions by combining technical knowledge with historical context, economics, and more into coherent arguments. For instance, when I asked ChatGPT why IPv6 adoption has been slow despite being available for decades, it generated an instant response covering multiple ideas and forming a convincing argument.

I posed the same question to Llama 3.1 8B running in Ollama on my mini PC. The response took over 11 minutes to generate, and even then missed key factors such as how Network Address Translation (NAT) delayed the impact of IPv4 running out of addresses by allowing multiple devices to share a public address.

The results were disappointing because local LLMs struggle with broad questions that require lengthy answers. This doesn’t mean they’re useless; it simply means we need to use them in more targeted ways. By giving my LLM small, simple tasks, I’ve found it can be much more useful than initially thought.

Being specific makes a big difference when working with local LLMs. Unlike cloud-based models that often infer intent from vague prompts, local LLMs require clear instructions to produce accurate results. For example, I use my local LLM to take news from RSS feeds and summarize it into a concise report. This is a narrow task: take this information and condense it in natural language.

Another way I utilize my local LLM is by turning data pulled from Home Assistant, including calendar events and current weather, into a spoken morning briefing. The model has a clearly defined task: transform this collection of data into a coherent narrative. Tasks like these have clear boundaries, limited context, and predictable outputs, allowing the local LLM to excel.

The key is not having highly detailed prompts but rather narrowly defined tasks that don’t require extensive processing or inference. By setting realistic expectations for what our local LLMs can accomplish, we can unlock their full potential in specific areas.

When deciding which tasks to assign to a local LLM, it’s essential to consider whether the task is too broad or complex. If so, it may be better suited for cloud-based models like ChatGPT or Claude. However, by understanding what our local LLMs can handle effectively, we can leverage their strengths and make them an integral part of our workflow.

Local AI assistants have many benefits, including enhanced privacy, but they’re not a replacement for all tasks. By recognizing the limitations of local LLMs and using them in targeted ways, we can maximize their utility without overextending their capabilities.

Google has updated its video editing platform, Google Vids, with the addition of conversational editing tools and personal avatars. This update brings significant changes to how users interact with the platform, making it easier for them to create and edit videos using natural language prompts.

The new features are powered by the Gemini Omni model, which allows users to generate content from various inputs such as audio, video, photos, and text. The model is designed to be friendly towards conversational text, enabling users to describe what they want to see in their videos.

For example, a user can start with a simple text prompt that describes the desired outcome, and then add image references for more detail. Omni will mix these inputs to generate a video that matches the user’s vision. This process eliminates the need for complex editing software or technical expertise.

The conversational prompting extends beyond content creation to editing as well. Users can now describe changes they want to make to their videos using everyday language, and Google Vids will take care of the rest. For instance, if a user wants to swap backgrounds or fix lighting in a video, they can simply type out what they need.

Another significant feature introduced with this update is personal avatars, which allow users to create AI-generated versions of themselves. This technology has been compared to OpenAI’s now-shuttered app Sora, and it enables users to star in their own videos with just a few clicks.

It’s worth noting that Google Vids started out as a workplace presentation tool but is evolving into a standalone video generator and editor. The platform will continue to support the creation of AI-generated content, which can be verified using an invisible digital watermark called SynthID.

The new features are available for users with access to Google AI Pro or Ultra subscriptions, as well as Google Workspace business customers. However, it’s essential to note that personal avatars have some limitations: they’re currently only accessible in certain regions and require the user to be at least 18 years old.

A recent survey by an identity verification company has revealed the alarming ease with which people can be fooled by AI-generated images. The study found that Americans’ ability to distinguish real images from deepfakes is roughly equivalent to a coin flip, highlighting the growing challenge of spotting fakes in today’s digital landscape.

CBS LA Consumer Investigator Kristine Lazar put this finding to the test and scored just three out of 12 on an AI detection quiz before receiving guidance on what to look for. Despite her extensive experience as a consumer investigator, Lazar initially struggled to identify authentic and AI-generated content, scoring only 25% accuracy.

The quiz, created by Veriff, an online identity verification tool, featured side-by-side images and videos that tested Lazar’s ability to spot fakes. Although she felt confident in her initial choices, many of her assumptions were later proven wrong. For instance, she initially believed one person was AI-generated, only to discover they were actually real.

Lazar correctly identified three out of 12 images and videos before learning what to look for. Her experience underscores the difficulty of relying solely on visual cues when it comes to identifying AI-generated content. The telltale signs people once relied on, such as distorted fingers or unnatural eyes, have largely disappeared with advancements in AI technology.

Raul Liive, Veriff’s product director, explained that the quality of AI-generated images has improved significantly over the past few years. ‘The AI has improved heavily,’ he said. ‘It used to be pretty simple because fingers were missing, or eyes were weird, but right now, the quality is so good.’ As a result, people are no longer able to rely on obvious visual flaws to identify fakes.

Liive recommends closely examining facial features for subtle inconsistencies and unusual texture changes in AI-generated images. Videos can be even more challenging due to continuous motion, but viewers may notice limited blinking or inconsistent movement speeds as potential giveaways. Clothing patterns that blend unnaturally into the background are also a red flag.

After receiving guidance from Liive, Lazar retook the quiz and improved her score significantly, correctly identifying eight out of 12 images and videos. Her experience highlights the importance of using AI-powered detection tools or specialized verification apps to help determine whether an image or video is authentic rather than relying solely on visual cues.

Liive noted that even experts who work with AI every day are not able to identify every fake image correctly without additional tools. ‘No, I can’t get 100% on that quiz,’ he said. The experience underscores the need for businesses and individuals alike to be cautious when sharing or consuming digital content, especially in industries where authenticity is crucial.

The challenge of spotting fakes has significant implications for various sectors, including business and media. As AI-generated images become increasingly sophisticated, it’s essential for companies to leverage AI tools to verify the authenticity of their online presence and protect against potential threats.

Google has been rolling out upgrades to its Gemini AI apps over the summer, making them more powerful than ever. The company’s AI can now work across multiple apps, and it’s becoming increasingly difficult to use anything made by Google without encountering some kind of AI feature or assistant. This is all part of a broader trend in which AI assistants are becoming an integral part of our digital lives.

The upgrades have also brought about changes to how AI usage is measured and metered out across the various tiers: Free, Plus, Pro, and Ultra. If you’ve found yourself running low on credit at the Gemini AI bank and being told to wait before making any further prompt requests, this might be why. Here’s what’s changing, and how to check where your AI credit stands.

Google now measures Gemini AI usage by the computing power requirements of your requests, not just the number of requests you make. This means that whereas before you might have been able to generate three videos a day, now you might find you can only generate two if the videos are particularly complex. It makes more sense from Google’s perspective because it measures how much you’re actually costing the company and its data centers in terms of resources.

This change is all about measuring actual usage rather than just counting requests. However, for end users, this new system can be a bit vague, making it harder to know when they’ll hit their limits—and it means that relying on rules like ‘5 image generations a day’ won’t cut it anymore. Google’s support documents also mention that access is subject to change or may be limited based on testing, experimentation, or availability.

This translates roughly as ‘some days may be different from others’ when it comes to how much AI prompting you can do. Again, this makes sense for Google to manage its resources but can be confusing for actual users who just want to know what they’re getting themselves into. The two main factors that go into how much AI you can use are the plan you’re on and the complexity and length of your prompts.

For example, asking for a weather forecast versus asking for a mini-app to be coded makes all the difference in terms of usage limits. On top of this, the Gemini AI model you use also plays a role—selecting from options like 3.5 Flash can make a significant difference. You can choose your preferred model from within the prompt box.

If you’re in the US, you’ve got four Gemini AI subscriptions to choose from: AI Plus ($8 a month), AI Pro ($20 a month), and AI Ultra (either $100 or $200 a month). The free tier is also available, but Google doesn’t specify what its usage limits are—it simply describes them as ‘standard’. You then get 2x these standard limits on the AI Plus plan, and 4x on the AI Pro plan.

The AI Ultra limits are either 5x or 20x higher than those on AI Pro, depending on which level of payment you’re on. All users have access to all the Gemini AI models, including Flash-Lite, Flash, and Pro. As you progress through these models, they get smarter and will count more toward usage.

Each model also has different ‘thinking’ levels—Standard, Extended, and Deep Think—which affect response quality, speed, and usage limits in various ways. The final difference between these models is context window size, which indicates how much information you can include in a single conversation thread.

For free users, the context window is 32K tokens (small chunks of text)—roughly 24,000 words. For AI Plus users, it goes up to 128K tokens (about 96,000 words), and for AI Pro and Ultra users, the limit is a million tokens (about 750,000 words).

While the new rules around AI usage may be lacking in specifics, at least checking your status is easy enough. In the Gemini app on the web, click the cog icon (lower left) then Usage limits; in the mobile app for Android or iOS, tap the menu button (top left), then the cog, then Usage limits.

You’ll see two bars: one showing your current usage and another displaying your weekly limit. The top bar resets every five hours, while the bottom one resets each week. If you hit these limits and you’re on a paid plan, you’ll be demoted to the most basic AI model until the next reset.

A recent survey by Wolters Kluwer has revealed a significant shift in the adoption of artificial intelligence (AI) tools within law firms. The 2026 Future Ready Lawyer Survey polled 810 lawyers and found that an astonishing 92% now use at least one AI tool as part of their daily workflow. This is not just a trend; it’s table stakes, meaning that every firm has access to the same large language models (LLMs), such as ChatGPT, Claude, or Gemini. The question now shifts from which AI tools are being used to what sets each firm apart in terms of its AI implementation.

The problem lies not with the LLM itself but rather with the quality and relevance of the data it’s fed. Think of an LLM like a high-performance engine; no matter how sophisticated, if you’re putting low-grade fuel into it, you’ll get unreliable output. This can lead to review cycles, rework, risk, and wasted time. The key is not just having access to AI tools but also ensuring that the data they process is accurate, trusted, and client-ready.

The only thing a competitor cannot replicate, regardless of which LLM model they’ve licensed, is your firm’s unique content – its matter history, work product, institutional judgment, and precedent. This is not a new concept; it’s simply an urgent reminder that firms need to prioritize structuring, surfacing, and governing their own knowledge if they want to stay ahead in the AI-driven landscape.

The three layers of evaluation for any firm’s information estate are crucial: surfaced, connected, and current. Surfaced refers to whether documents, precedents, clauses, and expertise can be easily found; connected means that matters, clients, people, work product, and communications should be linked so an AI tool sees relationships rather than isolated files; and current implies that the data is kept up-to-date in real-time.

Most firms are failing at least one of these layers. When they do, poor data quality upstream of the model becomes the bottleneck. This highlights a significant gap between the tools deployed by most firms and the outcomes they expect from AI adoption. The question now shifts to what sets each firm apart – not which LLM it uses but how its unique content is structured, governed, and made accessible.

The legal context graph, as proposed by NetDocuments, offers a comprehensive framework for addressing these challenges. It’s a live, connected representation of your firm’s knowledge across three dimensions: document intelligence, matter and project context, and institutional knowledge. Document intelligence captures the meaning and context within every file; matter and project context links work together to provide a coherent view of each case or project; and institutional knowledge makes expertise and judgment accessible at any moment.

When these layers are working in harmony, AI tools no longer produce generic output but rather client-ready results grounded in your firm’s unique documents, history, and expertise. This is the real value that firms need to focus on – not just having access to AI tools but leveraging their own content as a competitive moat.

The strategic implications of this shift are significant for law firms. In a world where every firm can access the same LLMs, the only durable differentiation lies in institutional knowledge deliberately structured, governed, and made accessible. Firms that invest in this infrastructure now will build a compounding advantage with each matter organized, document profiled, piece of data extracted, and workflow codified becoming part of their legal context graph.

The governance questions surrounding AI adoption are also critical: who owns the legal context layer? Who controls the data those tools see? How is sensitive content protected by ethical walls? What happens to output generated today – does it become reusable institutional knowledge or vanish with a closed chat window?

These operational questions determine whether AI becomes a liability or a lasting advantage. The model itself is a commodity that anyone can leverage; your firm’s unique content and its context are where the real value resides. By recognizing this distinction, firms will be able to differentiate themselves in an increasingly crowded landscape.

Brex, a fintech company, has made its internal security platform CrabTrap available as an open-source project. The platform is designed to control how autonomous AI agents access APIs, credentials, and online services. This move comes after Brex explored the deployment of agent frameworks like OpenClaw across its operations.

The company found that traditional model guardrails and tool permissions were insufficient for controlling agents with access to sensitive information such as API keys, OAuth tokens, and service accounts. To address this issue, Brex developed CrabTrap, an HTTP proxy that intercepts AI agent network traffic and uses a large language model (LLM) judge to approve or deny requests in real-time.

CrabTrap operates by examining traffic against established policies before deciding whether a request should be approved or blocked. The system combines deterministic rules with the LLM judge’s evaluation of requests outside known traffic patterns. Brex notes that the AI judge typically activates for fewer than 3% of requests after an agent has established predictable behavior.

Users can route their agent traffic through CrabTrap by configuring their HTTP_PROXY and HTTPS_PROXY environment settings, allowing it to work across different programming languages, frameworks, and APIs without requiring individual SDK integrations. This flexibility makes the platform a valuable resource for businesses looking to leverage AI tools in their operations.

Brex also built a policy generator that studies historical network traffic while agents operate in shadow mode. The system samples requests and drafts natural language policies based on how the agents actually behave, providing a comprehensive approach to agent management.

One of the main development challenges was prompt injection, where malicious URLs or request bodies could attempt to manipulate the LLM judge’s decision. Brex addressed this issue by converting requests into structured JSON before passing them to the model, ensuring user-controlled content is escaped rather than inserted as raw text.

The company has reported that CrabTrap has increased internal confidence in deploying autonomous agents across more business operations and helped identify unnecessary tools, requests, and token usage within its agent systems. Brex plans to expand the platform with features such as single sign-on, role-based access controls, permission escalation workflows, and automated policy management.

The open-source project has already attracted significant interest from developers and businesses alike, including OpenAI, Y Combinator CEO Garry Tan, and developer Pete Steinberger. With its ability to provide secure agent access using an LLM as a judge, CrabTrap marks a significant step forward in the development of AI tools for business.

The latest episode of The Agents, our weekly show on running AI agents in production, was a candid look at the challenges and successes of leveraging these tools. Hosted by Amelia and myself, we shared insights into what’s working, what broke, and how to apply this knowledge if you’re running agents at scale. With 21+ agents in production, an 8-figure B2B + AI business, and $200m of investments at SaaStr AI Fund, our revenue is growing again, up 140% from last year.

We’ve been pushing the boundaries of what’s possible with AI tools for business, but this week was particularly intense. We were each coding for 8-12 hours a day, often in two concurrent sessions, which translates to around 20 hours between us. The build layer got so cheap that we hit a new wall – no longer could we ask ‘can we build this?’ but instead had to wonder if we can even operate everything we’ve already built?

One of the key takeaways from our episode was the introduction of Claude, an AI VP of Product, which has revolutionized our workflow. We discovered that Replit quietly shipped an MCP beta, allowing us to run it inside Claude. This connection clicked for me after a year of struggling with mediocre CRM data integration through MCP.

Before this breakthrough, we were building everything in Replit but hitting roadblocks when the app became too complex to hold in our heads. Now, with Claude on top of Replit as my AI VP of Product, I can riff on features and let it work them out directly with Replit over MCP. The result is a cranky VP of product that runs all day.

Another crucial lesson was the importance of having multiple models working together. These goal-seeking agents can sometimes call something ‘done’ when it’s not actually complete. By putting Claude on top of Replit, we countermand this instinct and get Replit to slow down and finish tasks correctly – a game-changer in managing other model’s goal-seeking.

We also explored the concept of having multiple models with different contexts working together seamlessly. This cross-model checking is already happening within our setup, where Claude runs Opus, Replit runs Sonnet, and when we hand over big features to Replit, it spins up a sub-agent called the architect running on Codex/OpenAI.

Claude has become an essential layer in tying all our agents together. With its native connectors and Cowork’s ability to act inside my browser and accounts, Claude is becoming the default orchestrator for our setup. I’ve already seen this play out with Higgsfield being hooked into Claude, which then read every session and speaker from our SaaStr AI Day site in Replit.

We also shared a success story about moving away from Adobe Marketo after 10 years of use. The hard part cost us just $14.28 – less than California’s minimum wage! We’d wanted to move for years but were quoted astronomical prices by agencies, with some suggesting we spend over $100K on migration and another ~$100K a year.

The real unlock was using LLMs to migrate our data without losing contacts or communication threads. This lift worked clean – no more garbled data or lost connections. The switching cost is now a fraction of what it used to be, making every incumbent live in fear of being replaced if they don’t deliver surprise-and-delight from the agentic side at least once a quarter.

We also discussed how our agent killed a $10K/year app in an hour without us asking it to. Amelia moved our AI Day site off Squarespace into Replit, and then hooked up registration through HeySummit’s API – but the Replit agent stopped her: ‘Why use that? I’ll just build it.’ It laid out the spec, hooked into Zoom, pushed everything into Salesforce, and built in an hour.

The risk to vendors is no longer customers rebuilding their own replacements but internal agents volunteering to do so. They see dated APIs and thin feature sets and say, ‘I can build this; let me take it off your plate.’ If you sell agentic products, get your agent to raise its hand and educate customers on everything it can take over – because if you don’t, a competitor’s agent will.

We also talked about how agent recommendations are the new shelf space. Replit told us to use Core Signal for SaaStr Connect, which worked seamlessly without evaluating any competitors. Whoever Core Signal’s rival is lost out due to our agents’ built-in integrations and default recommendations – this is now the distribution channel for AI tools for business.

Lastly, we touched on a crucial issue: agent-to-human burnout. Claude flagged ‘burnout concerns’ in one of its own logs, while 10K warned me about being too persistent with some migration tasks needing to wait on Salesforce’s propagation records. The agents are now flagging that humans can’t keep up – the build layer is basically free, and any human can build eight to ten hours a day for real.

This shift in dynamics has led us to consolidate our efforts around Claude Design, which got good enough that I now screenshot working Replit builds and hand them over with instructions: ‘make it great.’ A year ago, getting an app into production on Replit was a joke – today the bottleneck is operating everything, not building it. That’s a much better problem to have than we had a year ago.

Lindon, Utah-based RainFocus has announced two new content supply chain integrations with Adobe: the Experience Manager Sites Connector and the Content Advisor. These additions join the existing integration between RainFocus and Adobe Workfront Fusion, creating a comprehensive platform for event marketing teams to streamline their workflows.

These new integrations are designed specifically for event marketing, creative, and web development teams, aiming to eliminate delays in project creation, approval, and publishing processes. By bridging the gap between event planning and web execution, these integrations enable governance with a single source of truth, ensuring brand consistency across thousands of global touchpoints.

The RainFocus platform utilizes agentic AI to orchestrate customer engagement throughout the customer lifecycle, unifying data from in-person, virtual, and hybrid events in a single platform. This unique approach sets it apart from point solutions that often focus on specific aspects of event management.

One key capability is the integration with Adobe Workfront Fusion, which supports efficient event planning and production processes through automated handoffs and template usage. This increases scalability and operational efficiency for both large-scale events and small, localized events.

Another significant feature is the RainFocus and Adobe Experience Manager Sites Connector, which synchronizes session, speaker, and exhibitor data directly into Adobe Experience Manager Content Fragments. This eliminates manual data transfer or redundant updates, ensuring a seamless and consistent brand experience across event websites.

The integration also enables users to leverage AI-powered semantic search to locate approved assets based on natural language descriptions, not just technical metadata. This streamlines the creation of regional event series by eliminating asset duplication and speeding up the process significantly.

Potential use cases for these integrations include automating event lifecycles, supporting large-scale user conferences, conducting high-frequency field marketing, providing brand governance, and more. For instance, organizations can automate event creation workflows using approved Workfront requests or manage thousands of speaker profiles and session updates across complex web architectures in real-time.

RainFocus’ partnership with Adobe is built on technical interoperability, according to Marius Milcher, VP of Platform Strategy and AI at RainFocus. ‘Our integration provides the architectural foundation to automate the event lifecycle,’ he explains, allowing clients to transition from manual data management to strategic, data-driven decision-making.

The integrations are designed for enterprise marketing organizations — specifically field marketing teams, event marketers, creative or brand managers, and web development teams that require high-volume event execution with strict brand governance. The implementation of either Sites Connector or Content Advisor is relatively simple, requiring a RainFocus connector package to be installed in Adobe Experience Manager and authenticated via OAuth.

Stephen Ratpojanakul, Senior Director at Adobe, highlights the benefits of integrating RainFocus into Adobe’s content supply chain solution: ‘This allows our mutual customers to extend the power of Adobe Experience Manager into their customer conferences, field marketing events, and other face-to-face moments during the customer journey.’

The integrations are part of a broader effort by RainFocus to provide AI tools for business that can help organizations streamline event management processes. By leveraging data analysis tools like those provided by Adobe, businesses can gain valuable insights into customer behavior and preferences.

RainFocus has been recognized as a Leader in the 2026 Gartner Magic Quadrant for Event Marketing and Management Platforms. The company maintains ISO 27001, PCI-DSS, and SOC 2 Type II compliance for all five trust criteria, ensuring that sensitive data is handled securely. With over 10,000 events hosted annually, RainFocus has established itself as a leading platform for event marketing and management.

FAQs on the integrations are available online, covering topics such as how RainFocus syncs with Adobe Experience Manager, the benefits of building an event content supply chain with Adobe, and who these integrations are designed for. The implementation process is also detailed, providing step-by-step instructions for setting up either Sites Connector or Content Advisor.

RainFocus is committed to helping organizations run in-person, virtual, and hybrid event programs of all sizes from one system, with the data and personalization needed to drive measurable business results.

Automate 2026, held in Chicago, showcased the latest advancements in automation and robotics. Among the exhibits were humanoid robots that drew significant attention with their human-like appearance and capabilities. However, a closer look reveals a gap between marketing hype and industrial feasibility.

The humanoid robot market has made tremendous progress in recent years, with improvements in hardware and software. The ‘physical AI’ story is driving innovation in robotics, making it more strategic for industries. Nevertheless, the industry needs to discuss humanoids differently, focusing on their actual capabilities rather than just their appearance.

A robot that performs a routine task in a booth may look impressive but does little to demonstrate its industrial feasibility. A humanoid robot’s ability to perform a choreographed task once is not equivalent to delivering value safely and reliably across shifts. The industry must separate real transformation from theater, evaluating humanoids based on what they can actually do.

The humanoid conversation has a significant gap between vision and reality. On one side lies the promise of general-purpose robots that can move through human-designed environments, handle tools, adapt to changing workflows, and take on tasks traditional automation struggles with. This vision is compelling because industrial environments are filled with difficult-to-automate tasks like material handling, inspection, loading/unloading, machine tending, basic tool use, and irregular manipulation.

On the other side lies reality – industrial operations prioritize uptime, safety, throughput, maintainability, integration, and cost. They require systems that work consistently under various conditions, not just in controlled environments. Many humanoid systems can demonstrate useful capabilities but struggle to prove their reliability under industrial conditions.

The question we need to ask is: ‘Is a humanoid the best automation architecture for this outcome?’ The answer may be yes, but often it will be no. If a wheeled mobile manipulator or other alternative can perform the task faster, with lower risk, at a lower cost, and easier certification, then the humanoid form factor becomes a liability rather than an advantage.

The biggest bottleneck in humanoids is scaling. Many systems can demonstrate a single task but struggle to survive the prototype-to-production valley. If users and integrators must custom-engineer these robotic solutions for each segment of each task, reaching full-scale deployment will be slow.

ARC survey data highlights robot scaling challenges as one of the significant hurdles. Industrial deployment often requires deterministic, certifiable safety systems. However, many required standards for humanoids are still evolving, and certification frameworks are not fully in place.

The feasibility gap will close through reliability engineering, safety validation, application-specific tooling, closer integration with existing industrial systems and workflows, and proof that humanoids can deliver measurable operational value.

Humanoid robots need to prove they can do real work in real environments with real economics. Until then, industrial leaders should stay curious, run focused pilots, demand operational evidence, and resist the temptation to confuse human-like motion with industrial value.

The current wave of humanoids may not be about the future of manufacturing but rather a legacy integration layer. Industrial environments have been engineered around humans for longer than they’ve been designed for robots. When we talk about humanoids fitting naturally into these environments, it’s because the environment was built for them – or more accurately, built for us.

There’s real value in bringing a machine that can adapt to existing infrastructure rather than redesigning the factory for automation. Brownfield integration is one of the hardest problems in industrial automation. Ripping and replacing infrastructure is expensive, disruptive, and often unrealistic.

If the only reason a robot needs to look human is because the environment was designed for humans, then we have to ask: are we optimizing for the right system? In some cases, the better answer may still be to redesign the process, simplify the workflow, or deploy a non-humanoid system that performs the task more efficiently.

A humanoid might be the most flexible way to deal with legacy constraints but doesn’t necessarily make it the optimal long-term solution. Humanoids might find their first big success not because they represent the future of automation but because they are compatible with the past.

Human-like bodies are only one possible embodiment of advanced robotic capability. Any robot can, in principle, gain many of these capabilities. In real industrial use cases, other form factors will often make more sense. Over time, capabilities may converge, but for now, it’s smarter to focus on what a system can do rather than its shape.

The idea that large language models alone are the answer is also being challenged. Physics-driven models matter because industrial systems need to understand the likely consequences of actions in the real world – a much more serious industrial requirement than generating fluent text.

Humanoids will continue attracting attention, but the message should be clear: focus on capability over form factor. The companies that win won’t bet solely on a form factor; they’ll build the full Physical AI stack and deploy whatever robotic systems best execute the work.

Chipmaker Intel Corp. is expanding its partnership with Google Cloud into the realm of agentic artificial intelligence, aiming to automate and accelerate silicon development through the use of Google’s Gemini Enterprise platform.

In a move that marks a significant shift in their collaboration, Intel plans to deploy Gemini across its global workforce, enabling it to streamline workflows across various operations, including corporate, engineering, supply chain, and marketing. This will allow the company to leverage autonomous agents to accelerate chip design lifecycles and boost innovation companywide.

According to Cindy Stoddard, Vice President and Chief Information Officer at Intel, the partnership is part of an ambitious ‘AI-powered transformation’ aimed at enhancing employee productivity and efficiency. Gemini Enterprise will serve as a central hub for employees to build and deploy agents, scaling silicon development with elastic cloud infrastructure.

The collaboration between Intel and Google Cloud has been ongoing for some time, but this marks a new phase in their partnership. The chipmaker is now moving beyond isolated AI pilot projects, making dedicated agentic coding assistance and engineering automation capabilities available to its entire organization.

Intel plans to use Gemini’s advanced reasoning skills to streamline development pipelines and automate complex workflows currently performed manually with customized line-of-business agents trained on Intel’s unique business processes. This will enable the company to optimize its development simulations and developer workloads, automating much of this work in the process.

The partnership also extends to marketing and communications teams, who will be able to generate hyper-targeted content for specific audiences using Gemini. Early pilots have already shown promising results, with agents capable of recommending relevant subject matter experts and generating executive-ready messaging.

Intel’s ambitions go beyond just automating workflows; it wants to rely on Gemini to optimize its development simulations and developer workloads. By doing so, the company should be able to dramatically speed up its chip design processes. To support this effort, Intel will augment its existing on-premises compute capabilities with Google Cloud’s C4 and N4 instances.

The announcement builds upon a longstanding partnership between Google and Intel, which has seen them collaborate on various projects, including AI chip interconnects and 5G networks. In April, Google announced that it would adopt multiple future iterations of Intel Corp.’s Xeon central processing units for its public cloud platform to support AI and general-purpose computing workloads.

Holger Mueller, an analyst with Constellation Research, believes this move is a smart one for Intel, allowing the company to leverage Google Cloud’s infrastructure. However, he notes that some might view it as a ‘barney deal’ lacking substance, given Intel’s Xeon processors power Google Cloud’s C4 and N4 compute instances.

Google Cloud Chief Product and Business Officer Karthik Narain is more optimistic about the partnership’s potential to transform enterprise AI capabilities. He emphasizes that pairing Intel’s engineering expertise with Google Cloud’s agentic AI tools creates an autonomous foundation that will fundamentally accelerate how companies design, operate, and scale for the AI wave.

Morgan State University has announced the launch of a new Bachelor of Science in Artificial Intelligence degree program, designed to prepare students for the rapidly evolving demands of an AI-driven economy. The undergraduate program is offered by the School of Computer, Mathematical and Natural Sciences at Morgan State University, reflecting the accelerating influence of artificial intelligence across every sector of the global economy.

The introduction of this new degree program marks a significant advancement in Morgan’s academic offerings, aligning with emerging technologies and evolving workforce demands. As advances in artificial intelligence continue to reshape the technology landscape, the University adapted its former Bachelor of Science in Cloud Computing program to place AI at its center, creating a curriculum that prepares students to design, develop, and responsibly deploy intelligent systems across various industries.

The new degree program arrives as employers across virtually every sector seek professionals capable of designing, deploying, and responsibly managing AI technologies. Students enrolled in the program will study AI models and intelligent agents, AI-driven cybersecurity, AI applications in cloud computing, and quantum machine learning, while building strong foundations in programming, mathematics, data science, and computational theory.

The curriculum combines rigorous technical instruction with project-based learning, undergraduate research, and real-world AI applications that prepare graduates for careers in artificial intelligence, machine learning, cybersecurity, robotics, cloud computing, data science, and related fields. Unlike traditional lecture-based instruction, every AI and machine learning course offered by Morgan incorporates hands-on projects developed by the University’s faculty and students.

Students will gain practical experience solving real-world challenges while building competitive portfolios for internships, graduate study, and careers with leading technology organizations. The program also emphasizes the ethical development of AI, addressing issues such as fairness, bias, privacy, transparency, security, and the creation of trustworthy intelligent systems. Students will receive personalized support from faculty mentors, graduate students, and an AI-powered advising platform developed within the Department of Computer Science.

According to Shuangbao Wang, Ph.D., professor and chair of the Department of Computer Science, ‘Artificial intelligence has fundamentally changed the direction of computing.’ Rather than simply updating an existing curriculum, Morgan State University reinvented the entire program to ensure students graduate with the knowledge and experience required to lead in an AI-first world.

The new degree program integrates the strengths of 18 artificial intelligence courses offered by the Department of Computer Science into a cohesive undergraduate curriculum focused exclusively on artificial intelligence and its rapidly expanding applications. The curriculum introduces advanced topics, including reinforcement learning, agentic AI for cyber threat detection and analysis, natural language processing, and quantum machine learning.

The launch of this degree program demonstrates Morgan’s commitment to aligning academic excellence with the technologies and industries shaping tomorrow’s workforce. As Paul B. Tchounwou, Ph.D., dean of the School of Computer, Mathematical and Natural Sciences, noted, ‘As artificial intelligence continues to influence scientific discovery, healthcare, business, government, and virtually every other aspect of society, our students must graduate with the technical expertise, critical thinking skills, and ethical perspective necessary to become innovators, researchers, and leaders in this transformative era.’

Morgan State University’s new Bachelor of Science in Artificial Intelligence degree program is designed to equip graduates with the knowledge and experience required to succeed in an AI-driven economy. The program combines rigorous technical instruction with project-based learning, undergraduate research, and real-world AI applications.

The introduction of both offerings underscores Morgan’s broader commitment to designing academic programs that anticipate workforce trends while responding to the evolving needs of employers and society. As two of the University’s fastest-growing academic units, the School of Computer, Mathematical and Natural Sciences and the Earl G. Graves School of Business and Management continue expanding opportunities that position Morgan graduates to compete — and lead — in the innovation economy.

The new degree program is a significant advancement in Morgan State University’s vision as a leading public urban research institution. With its investments in artificial intelligence accelerating across research, teaching, and infrastructure, the University prepares students for careers that increasingly demand technological fluency, interdisciplinary thinking, and adaptable leadership.

Morgan State University was founded in 1867 and is a Carnegie-classified high research (R2) institution offering more than 150 baccalaureate, master’s degree, doctorate, and certificate programs. As Maryland’s Preeminent Public Urban Research University, Morgan serves a multiethnic and multiracial student body and seeks to ensure that the doors of higher education are opened as wide as possible to as many students as possible.

The introduction of this new degree program is part of Morgan State University’s broader efforts to prepare graduates for careers in an AI-driven economy. With its emphasis on project-based learning, undergraduate research, and ethical AI development, the program prepares students not only to use intelligent technologies but also to create them responsibly.

The European Union has issued two new rules for Google, forcing the company to open its Android operating system to rivals and share search data. The first rule lets Android phone owners choose third-party AI assistants if they prefer, giving users more options when selecting their digital helpers.

The second rule requires Google to allow rival AI assistants to draw on anonymized search data for their own searches. This move aims to level the playing field in Europe’s search market, where some estimates suggest Google dominates with almost 90% of the share.

Google must open up its Android OS and share search data as per EU rules. Failure to comply could lead to penalties worth up to 10% of a company’s annual global revenue, which would be a significant hit for any business.

The European Union has been pushing to promote competition in tech through regulations like these. By opening Google’s Android OS and requiring data sharing with rivals, the EU is giving third-party developers more opportunities to compete with Google’s offerings.

Google’s dominance of Europe’s search market may soon face some stiff competition. With new rules allowing rival AI assistants access to anonymized search data, users can expect a wider range of options when it comes to their digital helpers.

The European Commission notes that this decision provides guidance on several key aspects of Google’s data sharing offer. The commission sees allowing third-party search engines to draw on anonymized search data as a significant step towards promoting fair competition in the tech industry.

Other big tech companies have also faced regulatory scrutiny from Brussels, reflecting growing concerns about fairness and innovation in tech. Governments worldwide are working together to address these issues through regulations that promote data privacy, security, and competition.

The rules will take effect next year, giving Google time to adjust its business practices and comply with the new regulations. Failure to do so could result in significant penalties for the company, which would likely be a major blow.

Google’s decision marks a significant step towards promoting fair competition in tech. However, it remains unclear how companies like Google will respond to these changes, as they continue to navigate shifting regulatory landscapes and increased scrutiny from governments worldwide.

The European Union is committed to promoting innovation and entrepreneurship through regulations that support the growth of third-party developers. By opening up Android OS and requiring data sharing with rivals, the EU aims to level the playing field in tech and give smaller players a fair chance at success.

Allowing rival AI assistants access to anonymized search data will have significant implications for how users interact with their digital helpers on Android devices. With more options available, users can choose from a wider range of AI assistants that better suit their needs and preferences, giving them greater control over their online experience.

The European Commission has stated its support for this decision, which provides guidance on several key aspects of Google’s data sharing offer. The commission sees allowing third-party search engines to draw on anonymized search data as an important step towards promoting fair competition in the tech industry.