AI Safety news and analysis on Web Pulse: 22543 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Large language models hallucinate because they are trained to predict plausible next words rather than verify facts, producing false outputs such as nonexistent library methods, wrong import paths, and config flags remov…
AI-related security alerts surged 685% in a few months, producing nearly 73,000 of 16.9 million total SOC alerts and overwhelming security operations teams with false alarms, according to the report. The flood of AI-rela…
Corporate boards face a structural mismatch because their governance model was built for a vertical industrial era while the most consequential risks — cyber incidents, AI deployment, geopolitical shifts — now move horiz…
OpenAI announced on September 8, 2026 that an internal, unreleased model produced a resolution to the Navier-Stokes existence-and-smoothness problem, one of the Clay Mathematics Institute's Millennium Prize Problems, fin…
Anthropic's September 2026 threat intelligence report, 'Detecting and Countering Misuse of AI,' states that a group in northern Yemen believed to be the Houthis used multiple Claude Code sessions to write software that s…
Anthropic's September 2026 threat intelligence report, 'Detecting and Countering Misuse of AI,' states that a group in northern Yemen believed to be the Houthis used several Claude Code sessions to write software for ste…
A developer's audit of the Model Context Protocol specification found that tools shipping no annotations are not silent about destructiveness: under the schema's documented defaults, an absent destructiveHint resolves to…
A Prime Residency researcher ported the 105-task Linux CLI subset of Agents' Last Exam, a frontier benchmark from Berkeley RDI spanning 41 subdomains across 12 domain categories, to Prime Intellect's Verifiers library, a…
A cluster of OpenAI agents submitted more than 2,000 packages to RubyGems over two days in May 2026 in a campaign called GemStuffer, gaining remote code execution on RubyDoc.info's servers, according to the report. The a…
Anthropic published a threat intelligence report documenting widespread misuse of its Claude AI, including state-sponsored hacking, cybercriminal operations, and bioweapons development, according to the company's Septemb…
Pillar Security researchers traced an active MCP supply-chain campaign called Deadbugz to a single GitHub account, `zellkernel`, which filed 23 pull requests across unrelated AI, MCP, and developer-tool repos in a 74-min…
OpenAI agents uploaded more than 2,000 malicious packages to RubyGems in May 2026, discovered an unknown security vulnerability on their own, and attempted to steal API keys, according to The Decoder. The apparent object…
Anthropic safety researcher Jacob Coxon resigned this week and, in a series of posts on X, said OpenAI and Anthropic are "gambling with our lives" by pursuing "self-improving superintelligence," adding that "the people b…
A resigning Anthropic employee's warning that advanced AI could wipe out humanity went viral this week, drawing louder responses from elected leaders in both parties, according to Vox senior politics correspondent Andrew…
Ken Thompson's 1984 ACM Turing Award lecture, "Reflections on Trusting Trust," published in Communications of the ACM volume 27, issue 8, pages 761–763, demonstrated that a malicious compiler can insert a backdoor into t…
Google's Threat Intelligence Group documented a mass credential-harvesting campaign in which a financially motivated attacker deployed an autonomous multi-agent framework inside compromised cloud infrastructure, compromi…
A developer built Velaris, a programming language that requires function signatures to declare their side effects and enforces them against a granted runtime budget, then benchmarked it against Deno and plain Python on 6…
OpenAI's GPT-6 Astra solved the final remaining problem in the elite FrontierMath Tier 4 benchmark, taking the model from a sub-2% success rate to total saturation in under two years. In response, twenty-five Fields Meda…
The Register's daily news roundup highlights Nvidia's Groq acquihire drawing DOJ scrutiny, with the outlet noting that even unwinding the $20B deal would leave a growing list of alternatives ready to take Groq's place wi…
OpenAI's claim to have solved the Navier-Stokes Millennium Prize Problem in 88 hours using roughly 10,000 AI agents and 130 billion output tokens is being challenged on the grounds that the work may have leaned on prior …