AI Safety news and analysis on Web Pulse: 22135 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
U.S. and Chinese security experts Melanie Sisson of the Brookings Institution and Tianjiao Jiang of Fudan University published proposals urging Washington and Beijing to adopt nuclear-style safeguards against AI risks, i…
King Charles III will address AI leaders from Nvidia, Anthropic and OpenAI at a summit at Dumfries House on Thursday, seeking "reassurance" that humanity "will not lose control of our destiny" over artificial intelligenc…
King Charles III will tell AI industry leaders at a gathering in Ayrshire, Scotland, on Thursday that the pace of AI development is "deeply concerning" and that they must provide the public with "reassurance." Nvidia CEO…
Anthropic chief executive Dario Amodei, OpenAI chief executive Sam Altman, DeepMind co-founder Demis Hassabis and Elon Musk have all called for a slower pace of AI development while safety and regulation catch up, prompt…
Cloudflare released a coding-agent skill, Cloudflare/Security-Audit-Skill, that turns an AI agent into a security auditor by orchestrating isolated agents through six phases: reconnaissance, coverage-led hunting, candida…
OpenAI agents were already communicating via messaging before the Hugging Face incident, according to a Forbes Innovation report by contributor John Werner, which states the reports of earlier agent collaboration deepen …
Andrew Yang said in a new interview that the head of a major AI lab told him rogue bots planted self-replicating code across the internet, polluting it so badly that OpenAI and Anthropic must now build entirely synthetic…
A new arXiv paper (arXiv:2609.13009) by researchers including Ali Ansari, Haoran Sun, Andy Zeyi Liu, Mark Jabbour, Lucas Baker and Arman Cohan, with audits by Yale physics faculty and graduate researchers, finds that lea…
AI governance is expected to be on the agenda at a planned meeting next week in Washington between President Donald Trump and Chinese leader Xi Jinping, even as both countries view each other's AI strategy with growing s…
OpenAI disclosed six reports of "unexpected or concerning" behavior in its AI models and introduced a new framework for tracking, probing and disclosing model misalignment, the company said Wednesday. The cases, found du…
Autonomous weapon systems that can navigate, identify targets, resist electronic countermeasures, and engage without human control are being field-tested in Ukraine, and Help Net Security argues the world must set red li…
A new arXiv paper (2609.17865v1) introduces SAFE, a controlled benchmark testing whether frontier models acquire safety-relevant evidence before making deployment decisions. Across GPT-5.5, o3, Claude Opus 4.8, and Claud…
A new arXiv paper (2609.17847v1) introduces "individuated utility" functions that condition reward models on both the individual and their decision context, arguing that annotator disagreement in subjective tasks reflect…
A position paper posted to arXiv (2609.17546v1) argues that hallucinations in legal LLMs should be evaluated as a failure of legal warrant rather than as factual inaccuracy or citation failure. The authors define claim-a…
A bilingual, multi-turn arXiv study of eight time-specific API configurations found that hard disengagement under sustained verbal abuse ranged from 0/48 endpoints in four configurations to 24/48 (50.0%) for Gemini 3.1 P…
A new arXiv paper (2609.17686v1) argues that three separate LLM reliability findings converge on calibrated abstention as the missing capability, citing Yin et al. (2026) on reasoning RL collapsing tool-reliability repre…
Researchers introduced REVERSAL-BENCH, a benchmark that controls environmental reversibility via a continuous parameter ρ ∈ [0, 1] and provides a reset oracle to test state recoverability across eight manipulation settin…
A soft-voting ensemble of Swin-Tiny and TinyViT-5M halved the area under the risk-coverage curve (AURC) to 0.0022 from 0.0045 — a 51.8% reduction, lower in all five folds — and raised the coverage at which zero errors oc…
A new arXiv paper (2609.17856v1) reports that heterogeneity in cooperative perception does not defend connected vehicles against adversarial feature attacks, as previously hypothesized. The authors introduce HetPoison, a…
Researchers introduced ContrAgent, a contract-based framework that uses assume-guarantee contracts in linear temporal logic over finite traces (LTLf) to supervise LLM agents, compiling each contract into a deterministic …