AI Safety news and analysis on Web Pulse: 9275 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
NVIDIA released a Secure Agent Workspace Reference Design to help enterprises govern autonomous AI agents that access sensitive data and business systems. The architecture shifts agent execution from user devices to mana…
A developer draws an analogy between carpentry and software engineering, arguing that complex systems like bridges and skyscrapers can hide fatal flaws beneath a polished surface, just as software can. The post cites the…
Prince Harry and Meghan Markle face backlash after claiming they were denied police protection for a UK trip, with journalists accusing them of emotional blackmail. Reports suggest Meghan has other reasons for avoiding t…
Blake Aber of Predicate Ventures argues that AI incidents in law firms, such as the Sullivan & Cromwell hallucination case, stem from infrastructure failures rather than policy gaps. He advocates for a 'harness layer' wi…
A philosopher argues that functional decision theory (FDT), favored by the Rationalist community, is both underspecified and implausible, making it definitely wrong. The author notes that almost no academic decision theo…
Blake Aber of Predicate Ventures argues that production AI success depends 90% on the 'harness'—observability, evals, rollback, and silent-failure detection—and only 10% on the model itself. He warns that enterprise AI p…
The Trump administration's export controls on Anthropic and OpenAI's frontier AI models are driving developers toward open-source alternatives, potentially benefiting foreign competitors. Commerce Department restrictions…
Bradley Scott Sayer, 18, was charged with two counts of first-degree murder in the Chico library shooting that killed two men. He faces life in prison without parole, and the court is not seeking the death penalty. Sayer…
TikTok creator Plonk is teaching viewers how to blind AI-powered surveillance cameras from Flock Safety, a network used by roughly one-third of U.S. law enforcement agencies. His satirical videos promote countermeasures …
The US and China are engaged in an AI-cybersecurity cold war, with open Chinese AI models advancing while US cyberweapons remain unavailable. South Korea plans a $590 billion investment in chipmaking facilities with Sams…
AI researcher and LessWrong community member argues that the concept of 'P(doom)'—the probability of catastrophic AI outcomes—is poorly defined and counterproductive. The post highlights ambiguity in what counts as 'doom…
A developer built LIME, a cryptographic identity layer for AI agents that provides signed JWT passports (RS256) verifiable by any website in under 10ms via a JWKS endpoint. The system uses Proof-of-Work challenges instea…
Bryan Cantrill, co-founder of Oxide Computer, argued in a talk that intelligence alone is insufficient for solving complex engineering problems, emphasizing that human values like resilience, teamwork, and optimism are c…
The integration of artificial intelligence into judicial and medical decision-making is accelerating, with systems like MIT's Sybil model detecting lung cancer years before symptoms and forensic tools entering expert rep…
The New York Times escalated its copyright lawsuit against OpenAI and Microsoft, accusing Microsoft of building a supercomputer specifically to help OpenAI steal copyrighted articles. The Times seeks to amend its complai…
Google has been secretly downloading its Gemini Nano AI model to eligible Chrome browsers since April 2026 without user consent, raising privacy concerns. The 4GB model powers on-device AI features and reinstalls itself …
BlackBerry shares surged after the company reported Q1 revenue of $152.9 million, up 26% year-over-year, and a net profit for the fourth time in five quarters. The former smartphone maker now focuses on its QNX operating…
A new report from Microsoft and Gartner reveals that technology teams are increasingly confident in deploying AI agents for tasks like generating reports and monitoring data quality, but confidence drops for complex task…
As recursive self-improvement accelerates, humans risk losing the ability to interpret and guide autonomous research agents, which could lead to safety failures. A new research agenda proposes studying how to keep humans…
Chinese AI company Zhipu AI claims its GLM-5.2 model matches Anthropic's Mythos at finding software vulnerabilities, calling it a 'cyber nuclear weapon.' Another Chinese firm, 360 Security Technology, also claims rival A…