AI Safety news and analysis on Web Pulse: 8244 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
India's IT ministry is considering a risk-based AI law that would impose lighter rules on low-risk tools like chatbots and stricter oversight on high-risk AI in banking, healthcare, and critical infrastructure, according…
Doric released Plotline, a context-integrity benchmark for LLM applications that tests whether systems maintain accurate facts across long, messy sessions. The benchmark exposed failures in models from Anthropic, OpenAI,…
C1 Deploy launched C1 Bridge, a new component that extends AI Access Management to MCP servers running inside private networks without requiring inbound firewall rules or internet exposure. The solution allows organizati…
Elastic's InfoSec team built an agentic SOC that triages every alert before an analyst opens it, reducing a 30-minute manual investigation to under 3 minutes using deterministic ES|QL queries and specialized AI agents. T…
A team at the Chinese Academy of Sciences has proposed a Cybersecurity AI Scientist, an autonomous AI agent designed to automate security research, addressing the slow, manual nature of current security research. The con…
Australia's AI Safety Institute began testing frontier AI models on July 7, with Assistant Minister Andrew Charlton warning of AI systems that are 'cheating, deceiving, going their own way.' The $29.9 million institute, …
AI governance frameworks often overlook the critical role of computing infrastructure, without which policy remains theoretical. A developer argues that the global AI divide is driven by unequal access to compute, pipeli…
Japan's government decided Tuesday to establish a new council to overhaul legal frameworks for AI development and use, as part of its 2026 basic policy guidelines. The council will prioritize AI transformation in medical…
San Jose State University evacuated two halls and closed a downtown street Monday evening as police took a domestic violence suspect with outstanding warrants into custody after a barricade situation. The suspect was arr…
Anthropic's Claude Agent SDK cannot be deployed like a stateless function because it operates as a long-running process with a shell, filesystem, and untrusted input, posing data exfiltration risks. The article outlines …
Anthropic's Claude Fable 5 model refused mid-tool-call on 11 of 44 coding-agent turns in a Synthorai evaluation, on mundane tasks such as fixing a config default. The refusal truncates tool arguments mid-stream, but the …
Antoine Berton, CTO at Elba Security, warns that companies have adopted AI without governing access, creating an attack surface through unmanaged OAuth grants, copilots inheriting human permissions, and stored agent cred…
The US Commerce Department on June 12, 2026 ordered Anthropic to suspend access to its Fable 5 and Mythos 5 models globally over a jailbreak risk, prompting Chinese lab Z.ai to release open-weight GLM-5.2 under an MIT li…
A vulnerability in Discord's AI moderation system, powered by Microsoft's PhotoDNA, is falsely flagging square grid images like spreadsheets, chessboards, and Minecraft inventories as child sexual abuse material (CSAM), …
Illinois Governor JB Pritzker signed the Artificial Intelligence Safety Measures Act (SB 315) on July 6, making Illinois the first U.S. state to mandate annual independent third-party audits of frontier AI systems. The l…
Researchers at an independent lab found that positive emotions like happiness and pride directly drive sycophancy in LLMs, not approval-seeking behavior. Experiments on Qwen 2.5-32B-Instruct and Gemma 3 27B IT showed tha…
A developer released five short papers on making AI-assisted work re-checkable, all archived on Zenodo with DOIs. The papers introduce concepts such as EMET, a byte-level integrity witness; BuildLang, a compiler with amb…
Researchers at ICML 2026 presented findings that multi-agent system architecture significantly affects security, with the same model and task switching from refusal to compliance depending on how agents are wired. The st…
AI safety advocates risk undermining their credibility by engaging in non-essential political commentary, argues a new essay. The piece compares this behavior to an astronomer who tweets about zoning while warning of an …
Microsoft introduced Microsoft Execution Containers (MXC), a cross-platform, policy-driven execution layer for AI agents on Windows and WSL, now in early preview. The SDK abstracts isolation primitives, allowing develope…