AI Safety news and analysis on Web Pulse: 10424 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Peking University's Chinalawinfo PKULaw has launched a standard Model Context Protocol interface that integrates large language models with authoritative legal databases, enabling AI-powered legal tools to generate accur…
Researchers introduced Yuvion VL, a family of multimodal large language models designed for content and AI safety, treating safety as an adversarial multimodal problem. The models employ a three-stage training pipeline a…
Researchers at arXiv found that in language models, the direction that best detects a behavior and the one that best controls it are often geometrically distinct, with cosine similarities as low as 0.12 for hallucination…
A new survey analyzes the use of large language models (LLMs) for scientific peer review, focusing on critique generation and score prediction. The study identifies reliability, robustness, and security risks such as pro…
Researchers at arXiv found that jailbreak attacks on large language models can be detected by analyzing entropy dynamics in intermediate layers, rather than final outputs. The study shows that monotonic rank-based trend …
A small group of Wikipedia editors, the Pro-Animal Wikipedians (PAW), measurably shaped how large language models discuss animal welfare through just 125 edits across 115 pages. Gradient-based data attribution and counte…
Researchers propose a test-time adaptation method using semi-supervised learning to improve AI text detection under continual distribution shifts, such as adversarial humanization and new LLMs. The approach leverages inf…
Researchers at arXiv propose a method to certify conservation laws in learned latent world models by decoding physical invariants from latent states. They derive shell-horizon certificates that decompose into representat…
Chinese AI models GLM 5.1 and Qwen 3.6 Plus, allegedly distilled from Anthropic's Opus 4.6, outperform the original on internal cybersecurity evaluations, including real-world vulnerability exploitation at a major US ban…
A new open-source tool, swarm-test, lets developers find failures in multi-agent AI systems before production by analyzing agent topology statically, without live LLM calls. It detects cascade failures, single points of …
White House negotiations with Anthropic over a restrictive export-control directive improved after CEO Dario Amodei stopped participating and co-founder Tom Brown took over, according to Wired sources. The directive, iss…
South Korean Second Vice Foreign Minister Kim Jina met with U.S. Under Secretary of State Jacob Helberg in Washington on June 24 to discuss AI and digital-technology supply-chain cooperation. The two sides agreed that tr…
A new benchmark, IatroBench, reveals that heavily safety-trained AI models cause iatrogenic harm by withholding critical medical information from patients while providing it to physicians, based on 3,600 responses across…
A developer introduced SQLazy, a tool that compiles natural-language step-by-step logic into auditable SQL, addressing the trust crisis in AI-generated SQL. Research shows LLMs achieve only 64.5% accuracy on Text2SQL tas…
A supply-chain attack on the @mastra AI-agent ecosystem saw an attacker republish 141 packages with a malicious dependency that targeted crypto wallets. The incident highlights how AI coding tools like Claude Cowork can …
A consensus is emerging on securing the agentic software development lifecycle (SDLC) as AI coding agents increasingly ship to production, but ready-to-use security solutions remain elusive. The number of pull requests h…
Anthropic accused Alibaba of running 28.8 million API queries through nearly 25,000 fraudulent accounts over 45 days to distill Claude's software engineering and agentic reasoning capabilities. The attack, disclosed on J…
Japanese railway operators are adopting AI-based systems to detect stalled vehicles or people trapped on level crossings, enabling faster response. Kintetsu Railway began full-scale operation of an AI camera system in Ma…
Anthropic's negotiations with the Trump administration to lift export controls on its Claude Fable 5 AI model have improved after CEO Dario Amodei was replaced by co-founder Tom Brown in talks, according to a Wired repor…
Scbkr, a local AI responsibility-chain control layer, introduces an owner-signed Workbench and Data Center with four-store evidence reuse to ensure models operate within user-defined rules and signed evidence, preventing…