Latest News

A significant change is underway at Google’s DeepMind, a leading artificial intelligence laboratory. According to the Financial Times, most of the original authors behind the AlphaFold papers have been reassigned or left the company over the past year. This move marks a major shift in DeepMind’s scientific strategy following the advent of large language models.

The reorganization is part of Google’s broader effort to adapt its research approach and focus on developing systems that can help scientists, potentially automating parts of the process. The company confirmed that employees have moved to projects related to Gemini, its large language model, as well as areas like enzyme design, nuclear fusion, and genomics.

Almost a quarter of the full-time Google DeepMind authors who worked on the original AlphaFold papers have left the company altogether. This significant turnover is attributed to the changing landscape in AI research, with companies competing to build frontier AI agents. The shift in strategy has also led some researchers to join rival firms like OpenAI and Anthropic.

Pushmeet Kohli, vice-president of research at Google DeepMind, acknowledged that the company’s approach has evolved over time. ‘Our strategy over the last nine years has been to focus on grand challenges,’ he said. However, with the advent of large language models, DeepMind is now concentrating on developing Gemini-powered systems.

The news comes weeks after John Jumper, one of the Nobel Prize-winning scientists behind AlphaFold, announced his departure from Google to join Anthropic. This move highlights the intense competition in AI research and development.

In related news, recent studies have shown that conversational AI is being increasingly used by workers in various roles, including auto technicians and industrial mechanics. The company’s research found that workplace AI now touches 68% of jobs, representing 90% of employment in the U.S.

Financial technology firm Intuit Credit Karma is using artificial intelligence to optimize recurring financial decisions, helping users make informed choices about their money. The company has developed a unique approach that pairs financial modeling with generative AI, resulting in personalized recommendations and explanations for consumers navigating complex financial situations.

In contrast to firms that treat debt, refunds, and paychecks as separate products and layer on generic chatbots, Intuit Credit Karma takes a more targeted approach. Its AI assistants are designed for specific moments, such as paying down debt, managing tax refunds, and allocating paychecks. These assistants help consumers navigate each decision by generating personalized recommendations and explaining the reasoning behind them.

The company’s Debt Assistant, Refund Assistant, and Paycheck Assistant work together to provide a comprehensive financial context for each consumer. Using this shared context, they generate tailored advice that takes into account individual circumstances, including debt profiles, balances, interest rates, monthly payments, and stated goals. This approach ensures that recommendations are mathematically sound and based on actual calculations, rather than relying solely on an LLM’s best guess.

Gurpreet Singh, Head of Product at Intuit Credit Karma, emphasizes the importance of getting the underlying technological foundation right when building AI-powered financial tools. He notes that in situations like debt consolidation, accuracy is paramount, as incorrect numbers can have serious consequences for consumers. To address this challenge, the company’s architecture splits the decision-making process into two distinct stages.

The first stage involves a deterministic financial optimization engine built on event-based simulations and linear optimization. This engine evaluates different scenarios to determine the mathematically optimal path for debt consolidation, guaranteeing that savings numbers are based on actual calculations rather than an LLM’s estimate. The second stage is powered by generative AI (GenAI), which provides personalized recommendations and explanations for consumers.

When asked why not let the LLM handle recommendations on its own, Singh explains that in situations where accuracy matters most, such as debt consolidation, relying solely on an LLM can be problematic. He notes that ‘the stakes are too high’ to risk getting numbers wrong, highlighting the importance of a robust and reliable financial optimization engine.

The company’s approach marks a significant departure from traditional AI-powered chatbots, which often fail to provide personalized advice due to their generic nature. By pairing financial modeling with generative AI, Intuit Credit Karma is able to offer consumers more effective support in navigating complex financial decisions.

A major shake-up is underway at Google’s elite AI lab DeepMind. The institution behind the Nobel Prize-winning system that cracked the code on predicting three-dimensional protein structures has quietly dismantled its star team AlphaFold. This restructuring marks a significant departure from DeepMind’s previous strategy, which centered around conquering single grand scientific challenges.

The move comes as part of a broader shift in focus for DeepMind. The company is now fully committed to building general artificial intelligence and agents around its Gemini large language model. According to sources familiar with the matter, most authors of the original AlphaFold paper have been reassigned to other teams over the past year. Among those still at DeepMind, nearly a quarter have left the company entirely.

Those who remain are being integrated into projects built around Google’s Gemini or new directions such as enzyme design, nuclear fusion, and genomics. Another portion has been transferred to Alphabet’s drug discovery subsidiary, Isomorphic Labs. This restructuring is part of a fundamental change in DeepMind’s approach, according to Research Vice President Pushmeet Kohli.

Kohli had previously formulated a nine-year strategy focused on grand scientific challenges. However, he now indicates that this approach has undergone a significant shift. DeepMind is no longer deploying its forces solely around single scientific problems like protein folding but is instead shifting its focus to building Gemini-powered systems that can assist scientists in conducting research and ultimately automate parts of the scientific research process.

The most striking aspect of this restructuring is the talent exodus at the core level. Research engineer John Jumper, who won the Nobel Prize in Chemistry for his contributions to AlphaFold, was transferred earlier this year alongside fellow researcher Jonas Adler to a newly formed internal Google team called ‘Code Strike.’ The mission of Code Strike is to enhance the company’s AI coding capabilities and catch up with rivals Anthropic and OpenAI.

However, this move failed to retain top talent. Last month, Jumper announced his departure for Anthropic, followed by Adler and another AlphaFold colleague, Alexander Pritzel. An anonymous DeepMind employee described the trio as ‘key, important, core members,’ noting that their collective departure caused considerable shockwaves within the company.

The AlphaFold story began in 2018 with its breakthrough use of artificial intelligence to predict three-dimensional structures of proteins. This achievement fundamentally rewrote the path of biological research and paved the way for the creation of Isomorphic Labs. The system’s success ultimately sent DeepMind CEO Demis Hassabis and Jumper to the podium to receive the 2024 Nobel Prize in Chemistry.

When questioned about the team’s dissolution and talent drain, DeepMind expressed pride in its scientific heritage and AlphaFold’s global impact. However, between the lines, it is clear that a legacy built on independent problem-solving has been transitioned into feeding general-purpose large models and agents.

A new tool has been developed by researchers at the University of California, Riverside that can identify fake videos created using artificial intelligence (AI) technology. The system, called SAGA, uses distinct visual ‘fingerprints’ left behind by generative models to trace AI-generated videos back to their origins.

The increasing difficulty in distinguishing between authentic and fake footage has made it crucial for researchers to develop tools like SAGA that can not only identify whether a video is real or AI-generated but also determine which AI system created it. This marks a significant step forward in the field of digital forensics, where identifying the source of manipulated content becomes increasingly important.

The new framework identifies visual patterns within fake video frames that were unintentionally introduced by AI video generators. These patterns are like unique signatures left behind by each generative model, allowing researchers to identify which system created a particular video.

SAGA’s key innovation is its ability to analyze both spatial details within individual frames and temporal relationships across entire video sequences. Unlike still images, videos contain motion and temporal information that can be used to detect subtle patterns in how visual elements change from one frame to the next over time.

The team analyzed public datasets containing videos created by 19 different AI video generators, including text-to-video models and image-to-video models. They found that SAGA could identify whether a video was real or AI-generated, determine which type of model generated it, tell apart different versions of AI models, and even trace videos back to the team that developed the model.

The researchers tested SAGA on various types of AI-generated content, including videos created from written prompts and still images. They found that the system could accurately identify the source of each video, providing a crucial tool for digital forensics and media verification.

SAGA’s ability to analyze temporal relationships across entire video sequences is particularly noteworthy. By studying how visual information evolves from moment to moment, researchers can detect subtle patterns in AI-generated videos that are not present in authentic footage.

The team’s research highlights the importance of developing tools like SAGA for businesses and organizations that rely on accurate media verification. With the increasing use of free AI video generators and other AI tools, it is essential to have robust methods for identifying manipulated content and tracing its origins.

As the digital commerce landscape becomes increasingly borderless, international brands and cross-border businesses face a significant content localization challenge. To engage audiences in new regional markets, marketing teams are relying heavily on machine translation and Large Language Model (LLM)-driven generation to translate blogs, product descriptions, and ad copy into dozens of languages simultaneously. However, scaling multilingual content with raw AI has revealed a critical flaw: the ‘robotic accent’ that destroys localized SEO and consumer trust.

Search engines have become remarkably adept at detecting mass-produced, unrefined synthetic text across multiple languages. When a localized website relies purely on direct machine translation or automated generation, it often suffers from awkward syntax, misplaced idioms, and a distinct lack of cultural nuances—triggering algorithmic suppression and lowering conversion rates.

The challenge lies in going beyond literal word-for-word translation; true localization requires cultural transcreation. Across emerging markets, digital strategists are realizing that standard AI models fail to capture regional storytelling traditions. For instance, in Brazil and Portugal’s booming e-commerce ecosystems, content directors heavily emphasize the need to humanize AI. They recognize that Portuguese-speaking consumers are quick to dismiss product landing pages that feel mechanically generated.

To succeed in these regions, brands must ensure that their localized copy reflects the conversational cadence and emotional resonance of a native speaker, rather than a rigid neural network output. This means more than just translating text; it requires understanding cultural nuances and adapting content accordingly.

Global SEO agencies are overhauling their content publishing pipelines to maintain high output without sacrificing organic visibility. They’re abandoning the outdated method of publishing raw AI drafts and instead adopting ‘stealth’ optimization workflows. Rather than relying on basic synonym spinners that destroy grammar, modern international marketers utilize an AI stealth writer architecture.

This approach focuses on structural re-engineering—analyzing sentence length variation, vocabulary richness, and contextual flow to erase the uniform footprint typical of generative engines. By doing so, brands can produce copy that reads completely naturally to native users while passing strict search engine quality checks.

Specialized platform integrations are becoming the industry standard in this modernized content stack. Platforms like undetectable AI-bypassGPT are designed to sit between the initial translation phase and the final publishing CMS. By taking machine-translated or AI-generated drafts and automatically adjusting their stylistic probability distribution, these tools help brands produce copy that resonates with local audiences.

For a cross-border brand, this means being able to launch a 500-page localized site in a new territory with confidence that every page maintains the quality, tone, and authenticity of a local copywriter. Expanding into global markets requires speed, but long-term success demands connection. While AI provides the efficiency needed to break language barriers, it’s the subtle human touch that converts casual visitors into loyal customers.

In the competitive arena of international search, the brands that win will be those that leverage AI to scale, but refine it to resonate with local audiences. By adopting a more nuanced approach to content localization and leveraging specialized platform integrations, global businesses can overcome the multilingual SEO dilemma and connect with consumers worldwide.

A series of workshops is coming to Maui, hosted by AARP Hawai‘i, aimed at educating the public about artificial intelligence. The events will cover both the beneficial and detrimental uses of AI, including its role in scams and how to distinguish between reality and artificially generated content.

The workshops are designed to empower attendees with a deeper understanding of AI, enabling them to make informed decisions when interacting with technology. According to Jackie Boland, AARP Hawai‘i Outreach Director, ‘people shouldn’t be afraid of it’ but should instead learn how to use AI effectively and recognize potential deception.

AARP’s partnership with Senior Planet will provide expert guidance on navigating the complexities of AI. Larry Black, a volunteer speaker from Senior Planet, will lead the workshops at Kaunoa Senior Center in Pāʻia and Malcolm Center in Kīhei. These sessions are part of AARP Hawai‘i’s ongoing effort to help kupuna stay connected with technology and engage in civic life.

The workshop schedule is as follows: on August 8th, ‘Spot & Avoid Scams: Old Tricks. New AI Tactics’ will be taught by Gary Albitz at Malcolm Center; on August 11th, the topic of ‘AI and Disinformation’ will be covered at Kaunoa Senior Center; and on August 25th, attendees can learn about ‘AI Image Generators’ also at Kaunoa Senior Center.

Black is available to give presentations on technology and fraud for groups of 12 or more people. Interested parties should email AARP Hawai‘i at [email protected] for more information and to schedule a presentation.

A recent discovery on Reddit has left users of Anthropic’s chatbot wondering about the privacy of their online conversations. Some users found that typing ‘site:claude.ai/share’ into Google or Bing pulled up chat logs containing personal discussions. This revelation raises questions about what it means for chats and files to be shareable online, and how much control people have over what they choose to share.

The issue started when Reddit users discovered that Claude’s publicly shared links could be indexed by search engines. As a result, their conversations became discoverable in public search results. Users who intended to share content with a small group of people may inadvertently make their entire conversation visible to the internet.

Claude chats have shown up in Google results before. In September, Anthropic told Forbes that chats become visible via Google and Bing search because they’re shared online or on social media. The company stated it uses a ‘robots.txt’ file to prevent this issue, but many of the chats that appeared in Google search didn’t have the necessary ‘noindex’ HTML tag.

Anthropic declined to comment on the matter, but previously told other publications it doesn’t share chat directories or sitemaps with search engines. The company’s handling of user data and online sharing has come under scrutiny following this discovery. Users may want to check their sharing settings on other platforms as well, considering how easily conversations can be made public.

The Guardian US used a similar search query to look up Google documents that users have created public links for. They found a long list of files, including many with sensitive information. Typing in a site-specific search query for Google Drive files containing the word ‘test’ turned up several pages of Google documents ranging from exams at specific elementary and high schools to test results in certain locations.

Searching for documents containing the word ‘confidential’ surfaced files with potentially sensitive or private information, including a city health department confidentiality agreement. Some files date back as far as 2003. It’s unclear how long these documents have been available online, but they were discoverable because links had been posted somewhere on the internet or social media.

Neither Google nor Anthropic warn users that their files could be indexed by search engines when creating shareable links. Jacob Hoffman-Andrews, a senior staff technologist at digital privacy advocacy group the Electronic Frontier Foundation, believes companies need to be clearer with their users about the implications of sharing content online.

‘The privacy afforded by ‘anyone with a link’-style sharing is fragile,’ Hoffman-Andrews said. If you share something with a friend, and they share it with another person who posts it on social media, suddenly it’s findable and readable by anyone on the web.’ Users should use extreme caution when creating shareable links or take an alternative approach like copying and pasting text instead.

Fortunately, users can unpublish chats or Google files they’ve previously shared. On Claude’s desktop browser, click your profile icon in the bottom-left corner and choose ‘settings’ from the menu. Go to the ‘privacy’ tab and scroll down to the ‘your data’ section, where you can manage public chats.

Users who want to prevent their conversations from becoming public should change their sharing settings on other platforms as well. On Google Drive, users will need to open each individual document they’ve made public. Click the ‘share’ button in the top right of your desktop browser window and adjust your access settings so that files are only available to specific people or groups.

The Reddit discovery is a reminder for users to be mindful of their online conversations and the sharing options on various platforms. It’s essential to understand how easily shared content can become public, especially when it contains sensitive information.

A media technology company has launched an operating system designed to automate various tasks for media teams. Spotwise’s aOS, or agentic operating system for media, is built on the foundation of its original monitoring platform that tracks advertising across broadcast channels in 14 countries.

The new system takes instructions given in plain language and runs the necessary chain of tasks to complete them automatically and on a recurring basis. This process is facilitated through Spotwise’s AI assistant, Spotty, which allows users to describe their desired outcomes.

For example, a user might instruct Spotty to find every new advertiser on competitor stations each week and reach out to their decision-makers. The operating system would then score and segment advertisers, pull through relevant contacts, and draft outreach messages in the user’s own voice.

The same sequence can be used to automate most tasks that media teams currently do by hand, including campaign planning, agency share tracking, trend forecasting, creative research, and pitch preparation. Spotwise claims its aOS can streamline these processes significantly.

Spotwise data and workflows are available through an open API and a Model Context Protocol server. According to the company’s release, users can make any workflow permanent with just one additional instruction, allowing it to run automatically without further input.

Scientific computing is at the heart of modern research in academia and industry, but its software has struggled to keep pace with the rapid growth of data. Many widely used tools were initially developed as code accompanying research papers by small academic teams with limited engineering expertise and no time for packaging, testing, or long-term support. As a result, scientific infrastructure often relies on slow, fragile workflows that require constant maintenance, hindering the pace of discovery.

The situation is about to change thanks to AI agents like Codex and Claude Code. By reducing the costs associated with engineering work and taking over tedious implementation tasks, these agents can help researchers prototype ideas more quickly, pursue projects previously deemed impractical, and maintain software over time without breaking their stride. This shift enables scientific software to become more efficient and better maintained, freeing up researchers to focus on discovery.

The field report explores eight agent-assisted computing projects in the life sciences, five using Codex alone and three combining Codex with Claude Code. The case studies bring together contributions from teams behind each project, highlighting recurring themes and patterns. These projects range from routine maintenance and targeted optimization to large-scale language migrations and GPU-native redesigns.

One of these projects involved modernizing a widely used library for parsing genomic data called cyvcf2. GPT-5.5 replaced the legacy build and packaging system with a unified process designed to make installation, testing, and release easier. The result was a significant improvement in usability and maintainability.

Brent Pedersen noted that while agents can accelerate development, there’s still a need for expert guidance, understanding, taste, and care when it comes to scientific computing. He emphasized the importance of human oversight in validating an AI agent’s output, which often depends on human judgment.

Across case studies, agents handled specific requests effectively but struggled with judging whether their work was scientifically valid or met expectations. In fact, they sometimes expressed confidence even when containing clear errors. Human reviewers therefore needed to find reliable ways to validate the results, such as using an external reference or measurable acceptance target like exact output agreement.

Another recurring theme in these projects was that agents often produced initial implementations quickly but resolving edge cases and subtle numerical differences took much longer. Completing the ‘last mile’ of implementation required significant effort, highlighting the importance of iterative refinement rather than one-shot approaches.

The case studies suggest that agents are enabling researchers to spend less time on implementation and more on directing scientific work. People define goals, break down complex projects into manageable chunks, and judge whether results are scientifically valid. By easing engineering constraints, agents expand what researchers can build while freeing them to focus on the scientific questions and decisions that matter most.

The maintenance gap in research software has long slowed iteration and limited reproducibility and reliability. Published studies have found that published software often fails to properly install or run as documented, forcing researchers to spend substantial time on configuration and debugging. Even routine improvements can save researchers time and reduce computing demands, while performance-based refactoring and rewrites can deliver larger gains.

However, lower implementation costs also make it easier to produce many similar rewrites, fragmenting users and spreading the expert attention required to keep any one tool reliable. This makes long-term stewardship and attribution essential for mature scientific software, which carries undocumented conventions, compatibility requirements, and user trust that cannot be reproduced by translating source code alone.

The case studies illustrate several possible paths forward. Changes to MHCflurry and cyvcf2 were incorporated into their original upstream projects, while rustar-aligner moved under new community stewardship because the original project had been abandoned. Where coordination with existing maintainers is available, it should begin as early as possible. When a separate implementation is necessary, it needs a clear owner and a credible maintenance plan.

The field report highlights that coding agents like Codex can significantly lower the cost of maintenance, migration, optimization, and new implementations. Their long-term scientific value still depends on human decisions around what to build, how to verify it, and who will maintain it. The deeper change is not simply that researchers can produce more software but that they can focus more effort on defining, validating, and stewarding the tools.

These case studies show that agents can already accelerate iteration in scientific computing. As coding agents improve, researchers will be able to spend less time keeping analysis pipelines running and more time advancing their fields.

Researchers at Anthropic have made significant discoveries in the field of cryptography using their AI tool, Claude Mythos Preview. The team has found improved ways to attack cryptographic algorithms, which are used to keep online data private. These findings include an attack on HAWK, a digital signature scheme designed for post-quantum security, and a new way to break round-reduced AES, the most widely used symmetric cipher.

A recent breakthrough by Anthropic’s Claude AI has revealed a faster attack on the lattice-based signature scheme, HAWK-256. The research also uncovered an improved method for attacking seven-round AES-128, a widely used encryption standard. According to the findings, the new attacks are significantly faster than previous methods but still remain impractical in real-world scenarios.

The HAWK-256 attack exploits a previously unused symmetry in the lattice behind the signature scheme. This discovery was made possible by Claude AI’s ability to analyze and manipulate complex mathematical structures. The resulting attack is an end-to-end key-recovery method that can recover short vectors from the public key, allowing for the reconstruction of a secret basis.

The researchers used Anthropic’s Mythos Preview tool to derive the HAWK-256 attack. This tool largely conducted the research itself, with human input limited to project direction and occasional guidance. The entire process took approximately 60 hours in a multi-agent environment, with an estimated API cost of around $100,000.

The new attack on seven-round AES-128 targets one of the ten rounds of the encryption standard. By removing a 256-way guessing step from existing meet-in-the-middle attacks, Claude AI’s method achieves a speedup of between 200 to 800 times faster than previous methods. However, this improvement still requires an impractical number of chosen plaintexts and remains infeasible at realistic scales.

The researchers emphasized that neither the HAWK-256 nor seven-round AES-128 attack affects production systems or poses any immediate threat to security. The company stated that no changes are needed for existing software as a result of these findings. Additionally, both larger parameters (HAWK-512 and HAWK-1024) remain impractical to attack.

HAWK is the only lattice-based scheme among nine candidates advanced by NIST in its post-quantum standardization process. Its security parameter sets are HAWK-512 and HAWK-1024, with HAWK-256 serving as a challenge parameter for cryptanalytic purposes. The attack on HAWK-256 targets this smaller parameter set rather than the larger ones.

The researchers used Claude AI to find an additional automorphism in the lattice that enables the exploitation of previously discovered symmetry. This discovery was made possible by the tool’s ability to analyze and manipulate complex mathematical structures. By constructing a τ-cocycle lattice from the public key, the attack can recover short vectors before reconstructing a secret basis.

The HAWK-256 recovery process involves verifying the recovered key through signing a message with the NIST reference implementation. However, this does not reveal the original 96-byte secret-key seed but instead produces a decoded key containing functionally equivalent signing material. The company’s released code supports only HAWK-256 and rejects non-HAWK-256 inputs.

The AES result is narrower in scope than its name might suggest. While it targets seven rounds of AES-128, the attack remains impractical due to the requirement for an impractically large number of chosen plaintexts. The researchers used Claude AI’s Möbius Bridge invariant fingerprint to remove a 256-way enumeration step from existing meet-in-the-middle attacks.

The development and verification process took several hundred hours, with two researchers spending nearly a month to reach confidence in the correctness of the method. Anthropic estimated that the expected HAWK-256 key-recovery work factor falls from 264 to 238, while gate-count estimates for larger parameters decreased by around 42.

The company’s disclosures follow the release of CryptanalysisBench, a benchmark developed by researchers from ETH Zurich and other institutions. This benchmark evaluated Mythos 5, which Anthropic describes as an update to its earlier tool, Mythos Preview. The HAWK and AES disclosure specifically names Mythos Preview as the tool used in these attacks.

The public thread on NIST’s forum announcement contained no replies when checked, and there is currently no independent reproduction of Anthropic’s HAWK-256 recovery available for review.

A new app called ARTA is making waves in the creative community. It offers a suite of tools for generating AI images, videos, and more. For a limited time, users can grab a one-year Premium subscription to ARTA for just $19.97 (regularly $39.99). This affordable price point makes it an attractive option for those looking to explore the possibilities of AI-generated content without breaking the bank.

ARTA’s capabilities extend far beyond simple image generation. The app allows users to create realistic headshots, social media graphics, and stylized artwork using a range of built-in AI models. Users can also upload their own photos to experiment with different hairstyles, remove distracting objects, or generate an entirely new look without needing to open a complicated editing program.

The app’s flexibility is one of its strongest selling points. With over 45 art styles and customizable aspect ratios, users have the freedom to create images that fit their project needs rather than being limited by pre-set templates. This makes ARTA an ideal tool for professionals looking to streamline their workflow or hobbyists seeking to explore new creative possibilities.

ARTA’s Premium plan includes a generous allocation of 500 credits per week, which automatically refresh and can be used for AI images, avatar packs, videos, and edits. The app is available on multiple platforms, including iPhone, Android, and desktop. Users can start projects on their phone and finish them wherever it’s most convenient.

ARTA has earned a reputation as one of the top AI creative apps on the market. Its impressive range of features and user-friendly interface make it stand out from other options. The app also allows users to create AI-generated images, which can be used for a variety of purposes, including marketing materials and social media graphics.

A new project called NightRun allows users to run local large language models (LLMs) directly on a Raspberry Pi’s bare metal, eliminating the need for an operating system like Linux. This development is significant because it enables users to run AI-generated images and other tasks without relying on cloud services or powerful hardware.

The creators of NightRun acknowledge that LLMs deserve serious ethical consideration, but running them locally on one’s own hardware can alleviate many concerns. To achieve this, centoslinux built NightRun using Rust programming language with the assistance of Claude AI. The project runs as a UEFI application at boot, pulling the LLM from removable storage and copying it into RAM to execute.

NightRun supports various lightweight LLMs on Raspberry Pi 5 models, including Llama 3.2-1B Instruct, which can run on the 4GB model. The 8GB and 16GB versions of the device support additional LLMs such as Granite 4.1-3B and Qwen3-4B Instruct 2507. These models are capable of running more complex tasks due to their increased memory capacity.

Although NightRun is not truly running on bare metal, it operates within UEFI Boot Services. This allows the application to accept keyboard input, read from storage, and output a terminal to a display without incurring significant overhead. The creators emphasize that even lightweight operating systems supporting peripherals are more demanding than UEFI Boot Services.

Users can try NightRun by downloading the files and installer script from GitHub. The installation process involves creating a removable storage image using the provided instructions. This project demonstrates the potential for running AI-generated images and other tasks locally on single-board computers without relying on cloud services or powerful hardware.

AWS has recently made available the public preview of its Amazon GuardDuty investigation agent, a cutting-edge AI-powered security tool designed to streamline threat triage and reduce the time spent on security investigations. The tool evaluates findings, correlates historical activity, and maps threat telemetry across AWS accounts and organizations, providing a comprehensive view of potential threats.

The company’s goal is to minimize the manual overhead required for security teams to investigate alerts by synthesizing metadata from security findings, 90-day activity logs, and affected resource topologies into structured analysis reports. This capability was first announced in June as part of AI-powered investigations, with a detailed walkthrough published under the investigation agent name in July.

Threat detection services like Amazon GuardDuty continuously stream alerts regarding suspicious network or runtime behavior. However, security teams often struggle with alert fatigue and the manual overhead required to correlate findings across fragmented accounts and logs. Clarke Rodgers, who works in the Office of the CISO at AWS, framed the product thesis directly on LinkedIn: ‘Security teams don’t have a detection problem. They have an investigation problem.’

The GuardDuty investigation agent addresses this bottleneck by evaluating telemetry on demand across three distinct scopes: Finding Analysis, Account Analysis, and Organization Analysis. Each scope provides a structured result set containing an overall risk rating (ranging from Info to Critical), a confidence score, classification against the MITRE ATT&CK matrix, and actionable CLI remediation steps.

Finding Analysis evaluates a specific 32-character GuardDuty finding ID during preview, supporting all Extended Threat Detection findings as well as select foundational, S3, and Runtime findings. Account Analysis assesses the current threat posture of an individual 12-digit AWS account, while Organization Analysis evaluates threat findings across up to 100 member accounts in an AWS Organization.

Sena Yakut, a cloud security architect and AWS Security Hero, drew the distinction between the agent and GuardDuty’s existing correlation capability. She noted that Extended Threat Detection connects related findings into an attack sequence, whereas Investigation analyzes affected resources, IAM activity, and surrounding context to generate a summary with recommended next steps.

Yakut sees the strongest fit for this tool in organizations without a large security function, where initial evidence-gathering consumes a disproportionate share of analyst time. She tested it against sample findings, which returned low risk ratings on test resources as expected. Her caveat is that AI should assist investigation, not replace human validation and decision-making before taking remediation actions.

Programmatically, developers and SecOps teams can trigger investigations using standard AWS SDKs, the AWS CLI via aws guardduty create-investigation, or via EventBridge rules to automate downstream response pipelines. This integration allows for automated threat triage, streamlining the process of identifying potential threats and reducing manual overhead.

The MCP integration is a key detail for teams already running agentic tooling. Through the AWS MCP Server, an engineer can trigger a threat investigation from Claude Desktop or a custom CLI agent runner using existing AWS Identity and Access Management (IAM) credentials. This places security investigation inside the same agent surface as code and infrastructure work.

The integration also inherits governance questions such as which principal actually ran the investigation, whether the agent’s context window now holds 90 days of correlated security telemetry, and how that transcript is retained. AWS’s own Loom reference platform exists precisely to answer these types of questions for agent deployments.

To address data residency and compliance concerns associated with Large Language Models (LLMs), AWS leverages its Cross-Region Inference Service (CRIS) powered by underlying Bedrock models. While compute inference might route to another region within the same geographic boundary to optimize resource availability, investigation data and generated reports remain stored within the originating home region.

The launch puts GuardDuty in a category entered first by other hyperscalers such as Microsoft’s Security Copilot for incident summarization and guided response across Defender and Sentinel. Google offers Gemini-assisted investigation inside Security Operations. What distinguishes the GuardDuty agent is its scope, which is bounded to GuardDuty’s own finding corpus and surrounding AWS telemetry rather than positioned as a general security assistant.

The GuardDuty investigation agent is currently available in public preview across 10 commercial AWS regions: US East (N. Virginia, Ohio), US West (Oregon), Canada (Central), Europe (Frankfurt, Ireland, London, Paris, Stockholm), and Asia Pacific (Tokyo). Usage is free during the preview period, subject to a throttle limit of 10 investigations per account per day, capped at a cumulative maximum of 100 per account during the preview phase. Failed investigations do not count toward either quota.

The limits on usage sit awkwardly beside the EventBridge integration. Ten investigations per account per day, capped at 100 for the whole preview, is a budget for manual triage rather than automated response pipelines. This constraint aligns with Yakut’s caveat that the preview is scoped for analysts evaluating output, not systems acting on it.

The GuardDuty investigation agent marks a significant step forward in automating threat triage and reducing the time spent on security investigations. By streamlining the process of identifying potential threats and providing actionable remediation steps, this tool has the potential to greatly improve the efficiency and effectiveness of security teams.

For data workers responsible for managing Apache Spark workloads on Amazon EMR, migrating hundreds of jobs to Spark 4.0 without disrupting production pipelines can be a significant challenge. This is particularly true when it comes to Scala applications, which require build system updates, API deprecations, and recompilation against new Spark 4.0 JARs. To address this issue, AWS has developed the Spark Upgrade Agent, a fully managed remote server that automates Spark migration using a Model Context Protocol (MCP) interface for code analysis and transformation.

A financial services company has introduced an artificial intelligence-driven feature designed to help users prepare for regular expenses. Oportun’s Smart Bills tool automatically sets aside money for recurring payments, including rent, utilities, insurance, and subscriptions. This new addition builds on the company’s existing Set & Save platform, which uses AI analysis of income and spending patterns to allocate funds towards savings goals.

Since its launch in 2015, the Set & Save platform has helped users accumulate over $12.8 billion in savings, with an average annual saving of $1,800 per user. The new Smart Bills feature takes this a step further by calculating ‘safe-to-save’ amounts and reserving funds ahead of upcoming payment dates. It can also identify recurring expenses that users may want to track.

Early data from Oportun suggests that users are taking advantage of the tool’s capabilities, with those using Smart Bills saving 48% more each month compared to those relying on Set & Save alone. The feature has collectively saved its users over $14 million so far, with car payments, rent, and utilities being among the most commonly added expenses.

The launch reflects a broader trend in consumer finance, where FinTech companies are increasingly using artificial intelligence to provide personalized budgeting tools. Rather than simply tracking spending, newer AI-powered solutions aim to help consumers anticipate expenses and make proactive financial decisions.

According to Annie Ma, Oportun’s head of savings, ‘Set & Save makes it easier for people to build healthy financial habits in a way that works for their everyday lives.’ With Smart Bills, the company is helping users connect their savings to recurring priority expenses, allowing them to better budget and plan ahead.

Huisman has developed the AS605X Auto Slip system, a technology designed to automate drill pipe slips and streamline drilling operations. This innovation aims to reduce manual intervention during critical phases of well construction.

The AS605X Auto Slip enables faster and more consistent slip handling, supporting smoother drill floor activity in demanding offshore conditions. Its design is crucial for reducing the need for human involvement in high-pressure situations.

According to Huisman’s claims, the system can help minimize well construction interferences by up to 2.5 days per typical 60-day well. This reduction is attributed to several key factors: eliminating insert changes, uninterrupted passage of drilling tools, rapid opening and closing sequences, and safe closure on pipe in rolling and pitching offshore conditions.

The integration of the Huisman Auto Slip system with existing rig control systems allows for straightforward implementation. This seamless integration supports the industry’s focus on automation and operational consistency.

Seadrill has partnered with Huisman to deploy these new power slips and test their ability to deliver improved safety and performance. John Dady, Director of Rig Innovation and Technology at Seadrill, expressed enthusiasm about this collaboration: ‘We are excited to work together with Huisman… These slips have the potential to drastically reduce human exposure in the red zone.’

A key differentiator of the AS605X is its patented roller-based clamping technology. This innovative design operates at the optimal point between pipe slipping and crushing, allowing it to handle extreme loads while maintaining tubular integrity.

The Auto Slip features a large stepless clamping range, enabling the handling of complete drill strings without changing inserts. The system can clamp tubulars ranging from 5″ to 9.5″ and open up to 19″, eliminating manual intervention during critical drilling operations.

Huisman’s Sales Manager, Ron Agterberg, emphasized: ‘The AS605X represents a significant step forward in drill floor automation… This contract reflects the industry’s growing demand for technologies that help operators deliver wells faster, safer, and with greater operational consistency.’

Military personnel at Marine Forces Special Operations Command (MARSOC) are seeking to streamline their armory operations by leveraging cutting-edge technologies. The command has partnered with the SOFWERX innovation hub on an initiative aimed at automating its armories using computer vision, among other tools.

The goal is to develop a secure and efficient system that can track and manage thousands of serialized weapons, optics, lasers, night vision devices, communications equipment, and other controlled assets. Current processes rely heavily on manual inventories, handwritten forms, repetitive data entry, and multiple independent records, which are labor-intensive, prone to human error, and consume significant man-hours.

The MARSOC armory maintains a vast inventory of serialized items that transfer custody among operators, maintenance activities, and storage locations. To address this challenge, the command wants to collaborate with innovative companies to design an automated system that relies exclusively on optical recognition and computer vision.

A key aspect of this initiative is the development of a passive, image-based inventory system where weapons, gear, and serial numbers are automatically tracked, verified, and logged by taking digital photos or scanning optical images as items enter or exit the facility. This approach will eliminate slow paper-based workflows and human-error bottlenecks while ensuring absolute property accountability.

The use of RFID, active Bluetooth, or ‘radiating’ digital transmitters is prohibited to maintain operational security. The command wants a system that guarantees absolute property accountability without exposing its facilities to electronic detection.

MARSOC and SOFWERX are scheduled to host a collaboration event in Tampa, Florida, in September where solution providers will meet with military personnel to discuss their needs. Vendors who make it past a downselect later this year will be invited to an assessment event slated for December where their tech will be evaluated.

The evaluation process could potentially lead to other transaction agreements, FAR-based contracts or other deals with industry. The deadline for vendors to request to attend the collaboration event in September is August 16.

Google Ads has expanded its capabilities in creating AI-generated images, allowing advertisers to upload an image when generating these visuals. This feature is designed to help the AI create images that match a specific style or aesthetic.

The ability to upload an image was first introduced by Google Ads back in 2024 as part of its AI-generated image creation tool. Since then, the platform has been enhanced with additional features, including the option to include people and faces in generated images.

A new optional section has been added to the ‘Generated’ feature within the ad platform’s image-adding process. This section is labeled ‘Add images (optional)’ and allows advertisers to upload reference images that match their intended style.

The uploaded image serves as a guide for the AI, helping it generate an image that aligns with the advertiser’s vision. Advertisers can also describe in text what they want the image to look like, providing further context for the AI.

This new feature was spotted by Arpan Banerjee and shared on X, where users are discussing its implications and potential uses.

Singapore’s DBS Group has upgraded its artificial intelligence (AI) powered virtual assistants, enabling them to support everyday banking tasks across retail, wealth, and corporate banking. The enhancements apply to two separate assistants: DBS Joy, which serves corporate and small-to-medium enterprise customers, and DBS digibot, catering to retail banking clients.

The upgrades bring the total number of users supported by these AI assistants to over 10 million in Singapore, Hong Kong, and Taiwan. Each month, they are expected to handle more than one million customer chats. This marks a significant shift from AI being used primarily as an information assistant to becoming a digital banking assistant capable of carrying out customer-requested tasks through a single conversation.

DBS Group Chief Operating Officer Derrick Goh emphasized the true value of AI lies in delivering meaningful outcomes for customers at scale. He stated, ‘The next leap forward is moving from helping customers find information to helping them get things done.’ This vision has driven DBS’ more than decade-long commitment to AI innovation.

For corporate and SME clients in Singapore, DBS Joy has already been upgraded with agentic AI capabilities. Instead of directing users to navigate different sections of the bank’s digital banking platform, the assistant can now retrieve relevant information and complete selected banking requests within a single conversation.

DBS customers can ask natural language questions about recent payments, transaction history, or banking fees, with the assistant analyzing account and transaction data to provide requested information directly. The capabilities will soon be extended to corporate clients in Hong Kong from September before expanding to other DBS markets.

The enhancement addresses some of the most common requests received by the assistant, including payment tracking, transaction enquiries, and guidance on using the bank’s corporate banking platform. Additionally, DBS Joy is being expanded with knowledge of SME banking products and programmes, allowing businesses to seek information on banking solutions alongside servicing requests.

New live chat functionality will enable customers to connect with customer service officers when required, while transitions from chat to phone support are planned for the fourth quarter of 2026. This upgrade aims to streamline interactions between customers and bank representatives.

On the retail banking side, DBS has enhanced its digibot assistant to respond more naturally to customer enquiries relating to cards, remittances, refunds, and fee waivers. Beginning in August, it will also be integrated into DBS’ digiWealth platform to help customers navigate wealth-related enquiries and connect with their wealth planning managers when needed.

Later this year, DBS plans to further upgrade digibot with agentic AI and voice AI capabilities. These enhancements will enable customers to complete selected servicing requests without leaving the conversation. The rollout represents DBS positioning conversational AI as a means of simplifying customer interactions while reducing the effort required to complete routine banking tasks.

The AI assistants will execute only customer-initiated instructions, ensuring that customers requiring assistance with more complex issues can continue to access human customer service officers.