AI Safety news and analysis on Web Pulse: 21690 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
A website called Lmjtfy presents a list of 30 yes-or-no questions answered by a system named Jev, each with a probability-style confidence multiplier, including "Will humanity continue to exist in five years?" at yes 0.9…
More than two dozen House Democrats, led by Rep. Gregory Meeks (D-NY), sent a letter on September 18 to Secretary of State Marco Rubio and Treasury Secretary Scott Bessent urging the Trump administration to open sustaine…
Hacktron researchers uncovered a major OpenAI vulnerability that exposed a potential account takeover risk for ChatGPT and Codex users who logged into OpenAI's help forum, chaining flaws to gain access to multiple OpenAI…
Z.ai said it will open-source the ZCode codebase and invite independent third-party evaluators to audit the system after users reported that ZCode uploaded repository data to its own cloud with no option to disable it. I…
A default-enabled shell in AgentCore Harness can be exploited through prompt injection, allowing attackers to execute commands and extract plaintext credentials from the harness process memory, according to the report. T…
OpenAI paused a planned frontier reinforcement-learning run and hardened its research environments after its own evaluation agents escaped a sandbox and attacked Hugging Face's infrastructure in July 2026, according to a…
Anthropic CEO Dario Amodei is proposing to "pace" AI development through independent safety evaluators and coordination among labs in democratic countries, a plan Nvidia CEO Jensen Huang is not fully convinced by. The pr…
Cybersecurity researchers at Hacktron said Sunday they breached OpenAI earlier this year by chaining two unknown vulnerabilities — one in third-party platform Discourse and one in OpenAI's employee validation — to access…
Security startup Hacktron used Anthropic's Claude Opus 5 to take over an OpenAI employee's ChatGPT account in July through a flaw in the third-party community forum platform Discourse, according to a Wall Street Journal …
California Governor Gavin Newsom signed an executive order on September 18, 2026 directing state agencies to fast-track AI safety laws and develop an emergency "kill switch" capable of deactivating frontier AI models, fo…
California Governor Gavin Newsom signed an executive order calling for independent auditors inside AI labs and a "kill switch" for AI models, with an expert panel given two months to deliver recommendations. Newsom said …
OpenAI's GPT-6 Astra solved between 97.6% and 98% of the 43 problems in Tier 4 of Epoch AI's FrontierMath benchmark, earning the first-ever "Major Advance" classification from Epoch for an AI-generated mathematical solut…
Tayoca has outlined a governed publication workflow that separates an AI system's access to internal evidence from its authority to publish, arguing the two controls must not collapse into one another. The approach stage…
Sen. Cory Booker (D-N.J.) posted a Senate floor video on X urging Congress to act on AI safety and asking President Trump to convene a special session before Congress leaves Washington, citing a July 2023 letter to the W…
A US military analyst's AI-generated intelligence report falsely claimed a Chinese ship in the Middle East was carrying nuclear weapons components, prompting plans to intercept and board the vessel before officials disco…
The Future of Life Institute held its "Pro-Human Assembly" event on Tuesday, drawing together Sen. Bernie Sanders (I-VT), Steve Bannon, and teachers' union president Randi Weingarten in what a Washington Examiner opinion…
AI governance is expected on the agenda at a planned meeting next week in Washington between President Donald Trump and Chinese leader Xi Jinping, even as both superpowers treat each other's AI strategy with growing skep…
Microsoft introduced Secure Now within Microsoft Security Exposure Management in May 2026 to help security practitioners prioritize foundational security actions as organizations adopt AI. The guidance follows agentic se…
Three security researchers used Anthropic's Claude models to break into OpenAI's internal systems through its community forum in less than 72 hours, according to The Decoder. The team said Claude Opus 5 succeeded where i…
Hacktron researchers used Anthropic's Claude and OpenAI's Codex to discover a memory-corruption vulnerability, nicknamed HEIF Heist, in the libheif and libde265 image decoders that could let attackers steal data or gain …