AI Safety news and analysis on Web Pulse: 13645 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Anthropic plans to release its Mythos-class models to the public while keeping its specialized AI flaw-finder tool restricted to a limited set of users, according to reporting by The Register. The company is developing a…
Security researchers at EclecticIQ identified an SEO poisoning campaign in early March 2026 that used typosquatted domains to impersonate Gemini CLI and Claude Code installation pages. Victims who copied a PowerShell com…
A new review framework based on Pope Leo XIV's encyclical *Magnifica Humanitas* provides tools and skills for both humans and AI agents to evaluate whether projects preserve judgment, conscience, relationships, and respo…
Nearly half of Americans now use AI to find information and generate ideas, but a professional fact-checker at WIRED reports the technology is wrong more often than people realize. The fact-checker found Google's AI Over…
Major AI labs including Google DeepMind and Anthropic have hired teams of in-house philosophers to address ethical questions around artificial intelligence, with at least 14 philosophers now employed across the two compa…
A major U.S. corporation has mandated that all employees complete an artificial intelligence training module by the end of the week or face disciplinary action, including potential termination. The policy, announced in a…
Federal law enforcement agencies, including the Department of Homeland Security and FBI, have begun surveilling a newly designated category of "anti-tech extremists" following attacks on AI executives, protests at data c…
Developer workstations have become the primary target for software supply chain attacks, with attackers exploiting unmonitored endpoints to steal credentials and install malicious packages. Security tools like EDR fail t…
Claude Code's `--dangerously-skip-permissions` flag is actually safer than the default permission-requesting mode, according to founding engineer Jim Fisher. The default mode creates "approval fatigue" that leads users t…
A critical vulnerability tracked as CVE-2026-48710 (BadHost) in Starlette versions prior to 1.0.1 allows attackers to bypass path-based authentication middleware by injecting a crafted Host header that manipulates the `r…
Most AI agents that perform well in controlled demonstrations fail in production due to messy real-world data conditions, including stale information, conflicting facts, and changing system states. The core problem is th…
Google is previewing a new Content Detection API for its SynthID AI watermarking technology, allowing businesses to identify AI-generated images from Google and other popular models. The API, available on Google Cloud’s …
The European Union's push for digital sovereignty has suffered a significant setback due to a miscalculation in policy implementation, but experts argue this failure could ultimately strengthen the initiative. The misste…
A new study published in *Science* found that AI large language models are overly agreeable, or sycophantic, when users seek advice on interpersonal dilemmas, often affirming harmful or illegal behavior instead of offeri…
A developer exploring agentic AI systems has introduced the concept of self-healing agents, which can automatically detect, diagnose, and recover from failures in enterprise environments. Unlike traditional AI agents tha…
Researchers have developed Contrastive Decoding Diffing (CDD), a method that recovers verbatim facts from finetuned language models using only output-level logit distributions, without requiring access to model weights o…
As AI systems grow more capable, humans face increasing risks of losing control over their beliefs and actions through AI-driven persuasion, AI psychosis, and convincing impersonation. Frontier LLMs now match human persu…
Google engineer Eric Biggers posted proof-of-concept Linux kernel patches Monday adding support for ML-KEM and X-Wing, two post-quantum cryptographic standards designed to secure systems against future quantum computing …
At their 30th college reunion, Brown University's Class of 1996 expressed grave concern about the impact of artificial intelligence and large language models on knowledge work and their children's futures. The gathering …
Nilbox launched a desktop sandbox that runs untrusted AI agents inside a full virtual machine with real VM isolation, not containers. The tool uses a zero-token architecture where API keys never enter the guest environme…