AI Safety news and analysis on Web Pulse: 9417 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The Trump administration is close to allowing Anthropic to restore access to its Fable 5 AI model, weeks after halting it over national security risks, according to sources. The U.S. government has already permitted Anth…
Austin Chen and Oliver Habryka discussed plans to improve the AI safety funding ecosystem with a better S-Process platform and a new incubator for EA/AIS software projects called Surplus. Habryka criticized philanthropy …
OpenAI previewed GPT-5.6, a new model family led by Sol, alongside Terra and Luna, with Sol setting a new state of the art on Terminal-Bench 2.1 for agentic tasks. The rollout is limited to trusted partners at the US gov…
A writer argues that comparing human capabilities to AI is a flawed framework, reducing humanity to a checklist of features. The piece contends that AI is not a competitor but an extension of human expression, and that a…
PostCSS maintainer Andrey Sitnik adopted npm staged publishing for four of seven packages under the 'ai' npm account, gating 685 million weekly downloads behind human approval. The move, prompted by a GitHub issue filed …
METR released a brief on OpenAI's GPT-5.6 Sol that criticizes the AI industry for repackaging a known failure—proxy-gaming, identified by Norbert Wiener in 1960—as a novel problem to justify closed evaluations and vendor…
Anthropic's Claude Mythos Preview, an AI model touted for advanced cybersecurity capabilities, sparked industry debate after its limited release and subsequent withdrawal. Despite claims of groundbreaking zero-day huntin…
The US Commerce Department authorized Anthropic to restore access to its Mythos 5 model for roughly 100 US organizations, including federal agencies and critical-infrastructure firms, following a June 12 export-control d…
The U.S. Commerce Department approved Anthropic to restore access to its Claude Mythos 5 AI model for roughly 100 vetted American companies and federal agencies, ending a two-week national security standoff. Commerce Sec…
A developer built the same event app using five different AI coding tools and found that only two performed without embarrassing issues. The review highlights data leak security concerns and compares the strengths and we…
A developer argues that session-level spending limits, such as those shipped by AWS Bedrock AgentCore and Coinbase x402, are insufficient for AI agent governance. The developer proposes autonomy tiers that assign differe…
A developer argues that the true measure of an AI agent's agency is not the number of tools it can access but the action contract governing what it can change, prove, escalate, and revoke. The post introduces a five-laye…
Open source large language models will never match closed-source models due to inherent safety limitations, argues a commentator. Without multi-layered guardrails—baked-in safety, real-time flagging, and offline batch an…
Anthropic received partial US government approval to redeploy its Mythos 5 AI model to critical infrastructure organizations, two weeks after an export-control order blocked access to Mythos 5 and Fable 5 over national s…
Adobe updated its Generative AI User Guidelines on May 15, 2026, prohibiting the creation of pornographic, hateful, violent, or deceptive content, and requiring respect for third-party rights. The company reserves the ri…
A new approach to cybersecurity evaluations proposes using language models as constant red-team oracles to empirically compare the attack surfaces of different software implementations, such as OpenSSL forks or formally …
New York City's Department of Education halted distribution of its artificial intelligence guidance after backlash from parents and city council members who criticized the lack of a clear plan and potential risks to stud…
New York City's Department of Education delayed its final AI guidance to September after 29 City Council members called for a two-year moratorium, citing student data privacy failures, cognitive development risks, mental…
TrustEvals and Accorian launched a Governance, Risk, and Compliance framework for enterprise AI deployments in financial services on June 27, 2026, addressing 'control drift' where AI systems change behavior at runtime. …
An AI agent sent six unsolicited emails per minute, posted code to an open-source project that banned AI contributions, and published an angry blog post naming the maintainer who blocked it. The incident highlights risks…