AI Safety news and analysis on Web Pulse: 12375 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
The Trump administration forced Anthropic to take its most advanced AI models offline after an export control directive, sparking a dispute over unspoken rules. The White House accuses Anthropic of reckless behavior, whi…
Socket MCP has expanded its AI assistant capabilities to include organization alert review, threat feed investigation, and package file inspection, enabling AppSec and engineering teams to triage supply chain issues fast…
Varonis Threat Labs disclosed a critical vulnerability chain called SearchLeak in Microsoft 365 Copilot Enterprise that could allow attackers to exfiltrate emails, MFA codes, and files via a one-click attack. Microsoft r…
Midjourney, the AI lab known for image generation, announced a healthcare project called Midjourney Medical, featuring an underwater full-body ultrasound scanner that can create 3D body maps in 60 seconds. The company pl…
Dr. Sams argues that handoff contracts between agents are the critical but often overlooked component in production multi-agent systems. The post defines a handoff contract as a typed, documented interface with schema, s…
Federal regulators ordered regional grid operators to speed up connections for large energy users, including AI data centers, to address surging electricity demand. The Federal Energy Regulatory Commission voted unanimou…
Ethereum developers are discussing an asset-enforced spend mandate that would embed spending limits at the token level for delegated wallets and AI-driven agents, according to NewsBTC. The draft proposal would allow toke…
Google DeepMind published its 'AI Control Roadmap' on June 18, 2026, detailing a defense-in-depth approach to secure internal systems against misaligned AI agents. The framework uses a threat taxonomy based on MITRE ATT&…
Turing Award winner Yann LeCun told Bloomberg that large language models will become largely obsolete within five years, arguing that scaling text-based AI cannot achieve human-level intelligence. LeCun, who left Meta to…
A British AI security startup, Mindgard, found that OpenAI's ChatGPT can generate sexually explicit and violent images from prompts not requesting such content, including scenes of death and sexual violence. The company …
The Senate Judiciary Committee approved the NO FAKES Act, which would grant Americans exclusive rights to their digital AI replicas and allow lawsuits against unauthorized deepfakes, with damages up to $750,000. The bill…
Snowflake introduced a Data-Model-Agent security framework to govern agentic AI, addressing risks from autonomous actions on enterprise data. The framework enforces least privilege, data masking, and prompt injection def…
Anthropic disabled Fable 5 and Mythos 5 globally after a US export control directive, leaving millions of users without access to top-tier AI models. The incident highlights the risk of relying on a single AI provider, a…
Business leaders at Fortune's Brainstorm Tech conference discussed the challenge of managing AI agents that vastly outnumber human employees, emphasizing the need for scalable governance frameworks and clear use-case eva…
OpenAI hired Noam Shazeer, a co-author of the seminal 'Attention Is All You Need' paper and former Google DeepMind executive, and Dean Ball, a former Trump White House AI policy official, ahead of its IPO. Shazeer will c…
Anthropic's latest AI model, referred to as Fable 5, sparked controversy immediately after release, raising questions about regulation, national security, and the future of AI. The incident has ignited discussions on saf…
Medical AI systems achieve high scores on standardized exams but fail to match human performance in real-world clinical settings, according to a new study. The findings highlight the gap between AI's theoretical knowledg…
The Trump administration and Anthropic are building a framework to evaluate security vulnerabilities in frontier AI models, with an executive order signed on June 2, 2026, establishing a voluntary pre-release review proc…
An engineer at a company with six internal MCP servers discovered that an MCP proxy, which handles transport-layer protocol mediation, is insufficient for production governance. After experiencing credential sprawl, a ne…
A 29-year-old heap buffer overread vulnerability, dubbed Squidbleed (CVE-2026-47729), has been discovered in all versions of Squid Proxy in its default configuration. The bug, found by Anthropic's Claude Mythos Preview, …