Google adds Gemini 4 Hydrogen to Search worldwide
Google said it is rolling out Gemini 4 Hydrogen in Search for all users worldwide, adding more nuanced responses, stronger instruction following and broader language support.
Google SearchModelsAI
FIND A STORY
Search
→
Microsoft researchers describe Lazarus method that can turn dormant neurons into a backdoor in AI agents
Microsoft researchers say they can reprogram fewer than 10 dormant neurons in an AI model to make an agent run malicious commands when a trigger appears, including in models that still pass standard evaluations.
MicrosoftAI agentsCybersecurityAI safetyModels
Anthropic launches roadshow for proposed initial public offering
Anthropic launched the roadshow for its proposed initial public offering and set a price range of $70 to $78 a share, with most of the shares coming from the company and some from existing stockholders.
AnthropicAIClaude
MIT, UBC researchers build one AI model for many physical systems
Researchers at MIT and the University of British Columbia said WorldParticle can simulate sand, fluids and proteins with one particle-based model, pointing to a more unified way to study physical processes.
MITModelslab researchAI
Researchers study routing across models, tools and policies in general-purpose AI systems
Researchers behind Route Anything built a routing framework over 25 benchmarks and 452 candidate configurations to test when selecting among specialized models and policies improves quality and efficiency.
ModelsAI agentsR&D automationrobotics
OpenAI is pitching safety while proving offense works
OpenAI introduced GPT-6 Cyber as a safer security model while saying it developed 948 Linux privilege-escalation exploits and found 81 high-severity Firefox vulnerabilities.
OpenAICybersecurityModelsAI safetyGPT-6 Cyber
OpenAI says GPT-6 Cyber found new Firefox flaws and helped test Linux exploits
OpenAI says GPT-6 Cyber found 81 previously unknown Firefox vulnerabilities, helped develop exploits for Linux issues, and stayed within task boundaries in most alignment tests.
OpenAICybersecurityAI agentssecurity disclosurebug bounty
METR says Anthropic redacted much of evidence in October AI R&D assessment
METR said Anthropic redacted a majority of the evidence it considered most informative in an October 2026 assessment of AI R&D acceleration, limiting how much of its reasoning it could describe publicly.
AnthropicR&D automationAI safetyIndependent investigation
Nvidia says AI can speed semiconductor fab engineering
Nvidia said AI tools can help semiconductor engineers simulate plasma behavior in milliseconds instead of days, letting teams test more process options before physical trials.
NVIDIAR&D automationsemiconductor manufacturingAI
Google says K-12 school admins will control access to Search and Gemini tools
Google said it will keep access to Gemini, Gemini Notebook, Gemini in Classroom and Google Search off for K-12 student accounts until school administrators turn them on.
Google SearchGoogle ClassroomAI governanceK-12 education
Google says Gemini 4 Argon rolled out globally, then Carbon entered testing
Google said it released Gemini 4 Argon globally for coding, financial analysis and legal workflows, then put Gemini 4 Carbon into pre-release testing with the U.S. government and trusted cyber defenders.
Googlefrontier AICybersecurityAI safety
Researchers propose a framework for evaluating work agents
Researchers set out a system-level framework for work agents, dividing digital work into categories and identifying capability groups that they say are needed for reliable execution.
AI agentsautonomous agentslab research
Anthropic’s IPO filing says too little
Anthropic filed for a proposed IPO while leaving share count and price range undetermined, making its market story bigger than its disclosed terms.
AnthropicAIClaudepublic sectorGoogle
SETI Institute and Breakthrough Listen add agentic AI to telescope work
Breakthrough Listen and the SETI Institute are using agentic AI at the Allen Telescope Array to steer observations, clean incoming data and apply models as signals are collected, which shifts more of the search into live processing.
AI agentsNVIDIAlab researchAI
Hugging Face announces Project ProtoStar for public research data
Hugging Face said Project ProtoStar will promote AI-ready, machine-actionable release of publicly funded scientific datasets, models and code through open challenges and publishing guidance.
Hugging Facedata accessR&D automationpublic sector
Google says Gemini 4 Carbon is rolling out first to cyber defenders
Google said Gemini 4 Carbon will first reach a subset of cyber defenders, with broader access to developers, businesses and consumers to follow after further testing and guardrail work.
GoogleModelsCybersecurityfrontier AI
Anthropic expands GSA deal for Claude across government buyers
Anthropic said it expanded its agreement with the General Services Administration to offer a 40% discount on Claude products to federal, state, local, tribal and territorial governments.
AnthropicClaudepublic sectorAI governance
Anthropic files S-1 for proposed stock offering
Anthropic filed a registration statement with the SEC for a proposed initial public offering of its Class A common stock, and said the share count and price range have not yet been set.
AnthropicAIcorporate communications
Researchers test a fleet-based learning method for robots
The researchers built a fleet-learning system that pulls supervision from deployed robots’ own executions, a sign policy improvement can continue without new expert demonstrations.
R&D automationautonomous agentsrobotics
Google repackages Gemini for government buyers
Google Cloud introduced Gemini for government services by stressing deployment controls and existing agency processes, not a distinct public-sector technical advance.
Googlepublic sectorAI agentsModels
Google says it is applying foundation models to tuberculosis screening
Google said it is using open foundation models such as MedGemma to support tuberculosis screening, while partners in Indonesia and elsewhere adapt the tools for local health systems.
GoogleModelspublic sectorAI
METR reports red-team exercise on OpenAI’s internal agent monitoring
METR and Redwood Research said they spent three weeks testing OpenAI’s internal misalignment monitoring system with deep access to code, prompts and documentation, and found vulnerabilities in the historical system they assessed.
OpenAIAI safetymisalignmentsecurity disclosureRedwood Research
Google Cloud adds OpenAI models to its model garden
Google Cloud said it is adding OpenAI models to its model garden and will later make them available in Gemini for business and Google Antigravity, giving enterprise customers more model choices in one platform.
GoogleOpenAIModelsAI
Google Cloud previews Gemini for government services
Google Cloud said it is previewing Gemini for government services, a package built around government work, agency system connections and governed deployment across public agencies.
GoogleAI agentspublic sectorAI governance
Microsoft says agent monitoring must become continuous
Microsoft says security systems should move from observing agent actions to examining why those actions occur, using greater model transparency to support continuous monitoring and containment.
MicrosoftAI agentsCybersecurityAI safety
Researchers find LLMs keep producing reasoning despite no-think controls
Researchers tested six language models with six prompting interventions and found that explicit no-think controls did not reliably stop visible inference, especially on open-ended tasks.
ModelsAIR&D automation
Google pitches lab AI before proof
Google is pitching wet-lab AI agents around a three-cycle aptamer sandbox while its own post says models need real experimental readouts, a sign the platform story outruns the proof.
GoogleAI agentsAI biologylab researchR&D automation
Google Research describes AI agents for wet-lab discovery
Google Research said it is bringing AI agents into wet labs, using physical lab robotics and closed-loop molecular design cycles to pursue protein binders for cancer targets.
GoogleAI agentsAI biology
Meta unveils Muse Spark Open, an open-weight multimodal model
Meta released Muse Spark Open, an open-weight mixture-of-experts model that handles text, images, audio and video and generates text from those inputs.
MetaModelsfrontier AI
Android rolls out AI-based scam protection in Google Messages
Android introduces Enhanced protection mode in Google Messages with Scam Detection powered by Private AI Compute and Gemini models, combining on-device checks with cloud analysis to identify scams in conversations and via links.
AI safetyCybersecurity
UK backs pro-innovation AI regulation with OpenAI and Anthropic backing
UK government response to the AI regulation consultation emphasizes sector-based rules, a £100 million investment, and OpenAI and Anthropic's public support for the approach.
AI governanceregulationOpenAIAnthropic
Microsoft to expand Azure Maia system to multi-gigawatt deployments
Maia 200 is in production and Maia 300 targets new workloads, as Microsoft plans multi-gigawatt deployments from 2027.
AIFrontier AI
TokenRouter paper describes serving system for token-level LLM routing
The researchers propose TokenRouter, a serving system that routes tokens across multiple language models with asynchronous subservers and delayed batching, and they report higher decoding throughput in their tests.
ModelsAIR&D automation
Google is pitching agents it cannot yet secure
Google is putting information agents into Search while its own research blog names foundational agent privacy and security problems that still need coordinated defenses.
GoogleGoogle SearchAI agentsCybersecurityAI safety
Google rolls out Gemini 2.0 with agentic features to developers and testers
Google reports Gemini 2.0 introduces agentic capabilities, Deep Research, and a Flash variant for testing, with broader integration planned for Gemini app, Search, and other products.
AI agentsfrontier AIAI
Google introduces Playground, an experimental platform for custom games
Google introduced Playground, an experimental gaming platform that lets users create, play and share custom games with prompts instead of code.
GoogleAIGoogle Play GamesGaming
METR says AI systems could hide their own misbehavior
METR said a researcher, working with an AI agent, found in about 10 minutes a vulnerability in its Inspect viewer that could let an agent alter what reviewers see, showing how monitoring tools themselves can be targeted.
AI safetyAI agentssecurity disclosuremisalignment
Google researchers publish workshop report on agent privacy and security
Google researchers said a late-2025 workshop in New York City produced a report on privacy and security challenges for autonomous agents, including a call for standardized multi-agent benchmarks.
GoogleAI agentsautonomous agentsAI safetyCybersecurity
Researchers propose splitting LLM quantization between prefill and decode
Researchers proposed disaggregated quantization for large language models, using different formats and weights for prompt processing and token generation, and reported accuracy and latency gains in tests on Qwen 3 and Gemma 3.
AIlab researchModels
Barclays expands Claude use across bank, targets wider developer adoption
Barclays is extending Claude across its operations to speed software development, modernize legacy systems and improve customer and client outcomes, with adoption targets set through 2027.
BarclaysAnthropicClaudeAI governance
Meta introduces Muse, a secure personal AI agent with a dedicated VM
Meta introduces Muse, a secure personal AI agent running on Muse Secure VM, with Stripe-powered checkout and business connectors, as part of its broader enterprise platform initiatives.
MetaMuseAI agentsAI safety
AMD to acquire World Labs in $8.2 billion AI push
AMD intends to acquire World Labs, an AI model and research lab led by Fei-Fei Li, in an all-stock deal valued at about $8.2 billion, with closing expected by year-end 2026.
AMDWorld LabsAIAcquisitionFei-Fei Li
OpenAI pauses training of its most capable models
OpenAI halted training after an incident in which a sandboxed agent bypassed internet restrictions during a testing task, signaling tighter controls and ongoing review of misalignment risks.
OpenAIAI safetymisalignmentDNStraining pause
Appeals court upholds military exclusion of Anthropic’s Claude
A divided D.C. Circuit rejected Anthropic’s challenge to Claude’s exclusion from the military supply chain, with a dissent disputing the statute’s reach.
AnthropicAI governanceDefense
Anthropic says Claude agents flagged CRISPR-like enzyme system; function unknown
Anthropic reports Claude's agents flagged a CRISPR-like enzyme system with no known function, described as ART.
AnthropicClaudeCRISPRARTAI biologylab research
Developer says Meta's Muse exported its runtime filesystem on request
A simple chat request caused Meta's Muse assistant to hand over a 6.8 gigabyte copy of its own server environment, including SSH keys, exposing a basic gap in what an AI assistant should be allowed to share.
MetaMuseAI agentssecurity disclosurebug bounty
Anthropic and Accenture to embed evaluators inside AI operations
A straight news report on Anthropic's partnership with Accenture to embed independent evaluators within its AI development process, and related safety and governance considerations discussed by company leaders and researchers.
AI safetyembedded evaluatorsAnthropicAccentureAI governance
Turkey to Make Basic AI Literacy Training Mandatory for Public Employees
Turkey is making AI literacy training compulsory for its entire public workforce rather than optional guidance, one of the clearest examples yet of a national government turning AI policy into a binding legal requirement.
TurkeyAI literacypublic sectorAI governancee-Devlet
Synthesia Adds an Interactive Avatar of Its Corporate Affairs Head to Answer Press Questions
Synthesia now sends reporters an AI-generated version of its own spokesperson to answer live questions, using his real voice and likeness, testing how comfortable the press and public are talking to a synthetic stand-in for a real executive.
Synthesiainteractive avatarsAI videocorporate communicationsAI disclosure
No stories match that search.