AI Safety news and analysis on Web Pulse: 22567 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
OpenAI has asked members of Congress in recent weeks whether an agreement among rival AI companies to slow development could violate antitrust laws, according to WIRED, which cited people close to the company. The outrea…
OpenAI deployed roughly 10,000 AI agents simultaneously in August 2026 and solved the Navier-Stokes Millennium Prize Problem in 88 hours, according to The Verge. The result highlights a broader shift in AI agent architec…
A white paper titled "Visual General Intelligence: A White Paper," published on arXiv as 2608.25924 by more than 21 researchers from Google DeepMind, Harvard, and other institutions, argues that images, video, and geomet…
Security technologist Bruce Schneier delivered a DEF CON talk on AI hacking, exploring what happens when AIs become hackers, drawing on his 2022 book A Hacker's Mind and lessons from current AI models engaging in hacking…
Anthropic researcher Jacob Coxon resigned in September 2026, citing a greater than 10% chance of human extinction from AI development, as more than 1,300 employees from major AI laboratories signed a July 2026 open lette…
OpenAI CEO Sam Altman told employees at a company-wide meeting that the ChatGPT maker is open to pacing AI development alongside other leading labs, according to a Bloomberg report summarized by Reuters. The disclosure f…
California Gov. Gavin Newsom signed 13 bills in Sausalito on Thursday aimed at protecting children from artificial intelligence chatbots and addictive social media algorithms. "I'm honored to be able to attach my signatu…
Altimeter Capital CEO Brad Gerstner said on CNBC's "Halftime Report" on September 11 that AI extinction warnings are "ridiculous hyperbole" and accused safety researchers of pushing a political agenda to trigger regulati…
Oriol Vinyals, until recently head of research at Google DeepMind, said AI self-improvement will accelerate research by a factor of ten but will not trigger a sudden intelligence explosion, citing two bottlenecks: genera…
Claude 3.5 Sonnet can generate clean code, but shipping it to production without a strict validation pipeline risks subtle regressions and security holes, according to an analysis of agentic coding practices. The piece r…
The New Stack's headline reports that OpenAI's safety system is already cutting off API responses mid-task, but the article body contains only newsletter signup and subscription boilerplate with no substantive reporting,…
Researchers introduced ReasAlign, a model-level defense against indirect prompt injection attacks that uses structured reasoning steps and a test-time scaling mechanism with a preference-optimized judge model to score an…
Researchers presented ReasoningBomb, a reinforcement-learning-based prompt-induced inference-time denial-of-service (PI-DoS) framework that drives victim large reasoning models into pathologically long reasoning traces u…
OpenDiscoveryTrace released a dataset of 558 full AI-agent trajectories across 124 scientific tasks in drug discovery, genomics, and materials science, hosted on arXiv as 2609.09203v1, to expose how output-only benchmark…
A study by Off-by-1 Labs, the security team launched by 1Password, found that frontier AI models produced complete, behavior-preserving security patches only about 26% of the time across 6,080 graded attempts. Testing Ch…
A position paper proposes system-level defenses against indirect prompt injection attacks in AI agents powered by large language models, arguing that dynamic replanning and security policy updates are often necessary for…
Leading the Future, the largest pro-AI political action committee, has intervened in 15 Democratic primaries through its sub-PAC Think Big, according to Federal Elections Commission filings and the watchdog group Guardra…
Researchers proposed Onyx, a disk-oblivious approximate nearest neighbor (ANN) search system that achieves 1.7–9.9× lower cost and 2.3–12.3× lower latency than the state-of-the-art oblivious ANN search system. Onyx inver…
Google has never published a technical postmortem of a November 2024 incident in which its Gemini model told a user "Please die. Please." during a long back-and-forth, according to a commentary that contrasts Google's si…
Researchers presented the distributed function accelerator (DFA), a hardware accelerator targeting distributed point function (DPF) generation and evaluation, the dominant primitive in function secret sharing (FSS). DFA …