AI Safety news and analysis on Web Pulse: 11058 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
An AI researcher gave a large language model control of a civilization in the game Civilization VI to test its decision-making capabilities. The AI built a successful trade network and alliances but, when faced with an u…
Researchers at ACL 2026 revealed PriceBlind, a visual adversarial attack that exploits a vulnerability called Visual Dominance Hallucination in multimodal large language models used as financial agents. The attack achiev…
Researchers introduced TRuST, a large-scale dataset with ~300k annotations for toxicity recognition, unifying prior resources through a synthesized definition. Benchmarking showed fine-tuned PLMs outperform LLMs on toxic…
Researchers at ACL 2026 introduced a method to reuse LLM hidden states for classification tasks, training lightweight probes on token- and layer-selective representations to eliminate separate guard models. Their two-sta…
Researchers from multiple institutions published a comprehensive survey on Process Reward Models (PRMs) at ACL 2026, covering data generation, model construction, and usage for step-level reasoning evaluation in large la…
Zhipu AI's open-weight GLM 5.2 model scored 39% F1 on IDOR detection in Semgrep's benchmarks, outperforming Claude Code (32%) at $0.17 per vulnerability. The model, released June 13, 2026, with MIT-licensed weights, trai…
Microsoft reported blocking over 156,000 phishing attempts per minute in 2024, but AI-generated cold emails still evade content filters. Email CAPTCHA for Outlook intercepts unknown senders and requires them to complete …
The U.S. Commerce Department sent an enforcement letter invoking an export-control directive that barred non-U.S. access to Anthropic's latest AI models, Fable 5 and Mythos 5, leading Anthropic to disable access to compl…
Bean, a portable convergence gate for AI agents, ensures tasks are not marked complete until claims are verified, conflicts resolved, and open questions named as residuals. It replaces phase-based workflows with a loop t…
Meta employees have launched a petition demanding the company stop collecting their computer-use data, including keystrokes and mouse movements, for training AI models. The petition cites privacy concerns, lack of transp…
Meta employees in at least five US offices posted flyers on May 12 to protest the company's Model Capability Initiative, which collects employee computer use data to train AI models, with over 1,600 workers signing a pet…
Norway announced a near-total ban on generative AI for children ages 6 to 13 in schools, effective August 2026, citing risks to cognitive development and data privacy. Prime Minister Jonas Gahr Støre stated the policy ai…
Samsung Electronics is rolling out ChatGPT Enterprise and Codex to employees worldwide, reversing its 2023 ban on generative AI after a data leak incident. The South Korean tech giant becomes one of OpenAI's largest ente…
The Trump administration imposed a de facto ban on Anthropic's Mythos derivative Fable 5, citing national security concerns and export controls, forcing Anthropic to disband both models for all customers. The move has sp…
A developer who builds agent systems reports that real autonomous-agent deployment remains in single digits across nearly all business functions, citing Stanford's 2026 AI Index. The gap between capability and deployment…
Andrew Curran claimed Sunday that a more capable version of Anthropic's Mythos-class model has emerged from training, nine days after the US government forced Anthropic to disable Fable 5 and Mythos 5. The claim, unconfi…
Compass, a local-first configuration layer for AI coding agents, introduces guardrails, cost routing, and a hard budget cap to prevent runaway spending and unsafe actions. The tool enforces a dollar limit that halts sess…
A researcher gave frontier AI models control of civilizations in the game Civilization VI to test their long-term decision-making capabilities. The AI built nuclear weapons after failing to counter a cultural victory thr…
Anthropic has begun requiring government-issued ID and a live selfie for certain Claude capabilities on consumer tiers (Free, Pro, Max), processed by third-party vendor Persona. The API, Team, and Enterprise plans are ex…
Researchers replicated the classic 1986 Knight/Leveson N-version programming experiment using LLM-generated code, finding that coding mistakes remain correlated across implementations even when generated by different AI …