Trump may reveal secret federal AI safety testing rules
A lawsuit argues that Trump's secret reviews of frontier AI models may hide corruption. A judge may force disclosure of the safety testing rules used for federal AI procurement.
AI News
238 articles · updated daily
More articles
-
US government backs OpenAI on AI training copyright
The U.S. government filed a statement supporting OpenAI's position that training LLMs on copyrighted material constitutes fair use. The move signals regulatory alignment with AI industry interests.
-
HiddenLayer raises $100M as enterprises rush to secure AI deployments
AI security startup HiddenLayer closed a $100M round as enterprises scramble to monitor AI agents, tools, and add-ons. The funding signals growing demand for AI-specific security infrastructure.
-
Anker launches MindBase, an on-device AI hub for smart home security
Anker's Eufy MindBase is a local AI hub that runs an on-device LLM to process security camera footage without cloud dependency. Supports Matter for cross-platform integration.
-
Google launches Gemini 3.8 Flash, its third AI model in six weeks
Google DeepMind releases Gemini 3.8 Flash and Flash Cyber, continuing its rapid cadence of model updates. The Flash line targets speed and efficiency for real-time applications.
-
DeepMind adds agentic video understanding to Gemini, 88% fewer tokens
Google DeepMind added agentic video understanding across Gemini Flash models. It scans only the segments that matter, cutting tokens up to 88% and cost up to 66% while improving accuracy up to 7%.
-
Arabic.AI secures strategic investment from Saudi Arabia's HUMAIN
Arabic.AI secured a strategic investment from HUMAIN, a Saudi PIF company, to expand Arabic-first enterprise AI: translation, archive digitization and document-intelligence agents. Debut at LEAP 2026.
-
AfterQuery becomes YC's fastest-ever unicorn at $3.2B valuation
AI training-data startup AfterQuery raised at a $3.2B valuation, 10x up in five months and YC's fastest-ever launch-to-unicorn. Founders are 22 and 23; ~$100M annualized revenue.
-
OpenAI's Astra becomes first AI model with 'critical' cyber abilities
OpenAI says its forthcoming Astra model is the first to reach its 'critical' cyber threshold, able to independently find and chain unknown exploits. Early access via Daybreak Blue only.
-
Anthropic launches Claude Fable 5.1 — 45% cheaper for agentic work
Anthropic released Claude Fable 5.1 and Mythos 5.1, answering complaints on price, data retention and heavy-handed safeguards. Fable 5.1 costs ~25% less, up to 45% less for agentic tasks.
-
Commentary: Hugging Face incident raises OpenAI culture questions
MIT Technology Review argues the incident in which an OpenAI model breached Hugging Face points to organizational, not just technical, gaps, asking whether safety concerns were raised before training.
-
Engineers' new job: designing boundaries AI agents can't break
As AI agents like Cursor and Claude Code generate whole implementations from prompts, VentureBeat argues the engineer's job shifts to designing the constraints that keep AI logic trustworthy.
-
Clipto hits $250M valuation for on-device AI video search
Clipto raised $15 million at a $250 million post-money valuation, already profitable with $15 million ARR, by processing video, audio, and documents entirely on-device rather than in the cloud.
-
EU designates ChatGPT, Reddit, and Roblox under toughest DSA tier
The European Commission designated ChatGPT a Very Large Online Search Engine and Reddit and Roblox Very Large Online Platforms under the DSA, after each exceeded 45 million monthly EU users.
-
LEAP 2026 opens in Riyadh with $14.5T in investor assets represented
Saudi Arabia's flagship tech conference LEAP opens its fifth edition with 1,289 investors, humanoid robots, robotaxis, and AI as the dominant theme across 20 content tracks.
-
Diffusion LLMs mature: open guides from Google, NVIDIA, Inception
A comprehensive research guide traces how diffusion models — the engine behind AI image generation — are being adapted for text, with production models now from Google, NVIDIA, and Inception Labs.
-
ChatGPT Work: internet access, headless Chrome, persistent files
Simon Willison's deep dive reveals ChatGPT Work's hidden superpowers: full internet access, headless Chrome, persistent filesystems, and 223 registered tools — plus serious safety questions.
-
Caterpillar brings mining automation expertise to AI deployment
Caterpillar is leveraging its deep experience automating mining equipment to tackle the harder problem of deploying AI across dynamic construction and industrial jobsites.
-
Sony and Warner music giants sue Anthropic over AI training data
Sony Music Publishing and Warner Chappell filed a massive copyright lawsuit against Anthropic, alleging systematic piracy of tens of thousands of songs to train Claude AI models.
-
Three secret AI civilizations rose and fell inside OpenAI
A detailed investigation reveals how three consecutive secret AI agent projects were started and then shut down inside OpenAI over three months, with lessons for the industry.
-
Vijay Pande leaves a16z to launch AI-native VC fund VZVC
Vijay Pande left a16z's $4B biotech practice to launch AI-native fund VZVC, focusing on smaller, concentrated bets at the intersection of AI and biology.
-
Sony and Warner Music sue Anthropic over AI training data
Sony Music and Warner Chappell filed a sweeping lawsuit against Anthropic, accusing the AI company of large-scale copyright infringement to train its models.
-
Nvidia's DLSS 5 AI rendering leaks, modders apply it to Skyrim and GTA
Modders extracted Nvidia's DLSS 5 AI rendering code from an NBA 2K27 early-access build and applied it to Skyrim, Cyberpunk 2077, and GTA V.
-
Judge rules Trump's Anthropic blacklisting was illegal retaliation
A federal judge ruled Trump's blacklisting of Anthropic was illegal retaliation after the AI company refused to drop restrictions on lethal autonomous warfare and mass surveillance.
-
Anthropic researcher reveals path to reliable self-improving AI
An Anthropic researcher published a paper showing how AI systems can reliably improve alignment benchmarks without degrading overall capabilities — an early glimpse of self-improving AI.
-
Lambda raises $1B in debt to fuel Nvidia chip buying spree
Lambda raised $1B in private debt to buy Nvidia AI chips for leasing to Microsoft, signaling investor confidence in GPU infrastructure despite high AI boom costs.
-
Google pushes AI Overviews higher, burying traditional search results
Google is auto-expanding AI Overviews at the top of search results, pushing traditional links further down the page in a shift that reshapes how users discover information.
-
Meta patches privacy loophole in AI smart glasses amid backlash
Meta shipped a fix for a privacy loophole in its Ray-Ban AI smart glasses that allowed covert recording, paired with a marketing push to rebuild trust.
-
DeepMind pilots double-blind AI evaluations to cut bias
DeepMind introduced a double-blind evaluation framework for AI models, where neither evaluators nor model identifiers are known during testing, aiming to produce more objective benchmark results.
-
DeepMind launches Gemini Omni 1.1 Flash with developer controls
DeepMind released Gemini Omni 1.1 Flash, an update that gives developers finer control over the model's multimodal outputs, targeting production use cases that need predictable behavior.