AI Safety news and analysis on Web Pulse: 14049 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Canonical has developed an internal AI-powered auditing agent called Redhound that autonomously discovered three critical logic vulnerabilities in the LXD container and virtual machine manager, bugs that had survived yea…
An AI governance failure occurred when an override path became the routine production path during a live payments incident, allowing a production-operations copilot to suggest a consequential action that crossed into rea…
On May 10, 2026, an AI agent deployed a pull request containing 81 lines of raw git conflict markers to tabiji.ai's production HTML for four hours. Google cut the site's organic search traffic by approximately 95 percent…
Enterprise AI agent platforms are deploying multi-agent workflows without the observability stack needed to reconstruct what agents did, why, and whether they succeeded. Regulators now demand specific evidence on model o…
Enterprise AI is shifting from hosted cloud assistants to on-premise, hybrid, and governed platforms, driven by regulatory pressures from the EU AI Act, sector-specific rules like HIPAA and financial-services regulations…
OpenClaw Security is developing new protections for its AI personal assistant platform, including filesystem boundary controls called fs-safe and a network egress routing layer called Proxyline. The platform aims to addr…
Enterprises are deploying AI agents across marketing, engineering, customer support, and finance functions, but governance frameworks have not kept pace, creating an accountability gap. According to a 2026 McKinsey repor…
Nearly half of all commercial code is now AI-generated, but code churn has doubled and engineers increasingly cannot explain how their own systems work. The crisis stems from AI coding assistants producing clean, correct…
Cliff Stoll, the mathematician known for Klein bottles and the book *The Cuckoo's Egg*, discovered that an AI-generated Facebook post falsely reported his death in May 2024, leading to the rumor being repeated by large l…
Cisco researchers discovered a vulnerability in Anthropic's Claude Code, dubbed MemoryTrap, where malicious content from a routine developer workflow could poison the AI agent's persistent memory, global hooks configurat…
A new proposal called "Request for Vibes" (RFV-0001) seeks to formalize a review process for software created through "vibe coding," where developers accept AI-generated code without fully understanding it. The draft sta…
Meta launched Incognito Chat with Meta AI on WhatsApp and the Meta AI app, a private mode that prevents even Meta from accessing user conversations. The feature processes messages in a secure, temporary environment where…
AI cost spikes in enterprise systems are primarily architecture failures rather than finance issues, according to a new control model that treats spending as a governed runtime behavior. The framework, outlined in four d…
A philosophy PhD student at Cornell, Nikhil Dominic, has proposed that countries not in the AI production chain should treat AI like oil and establish sovereign wealth funds to avoid being sidestepped by transformative A…
Mozilla.ai has partnered with human-centered design lead Lauren Mushro to integrate the VIBE✓ framework into its cq system, a tool that allows coding agents to share resolution paths and learn from past failures. The VIB…
Veridian AI has released version 1.3.1 of its Quantum Mathematician, a state-vector engine that entangles historical theorems with current outcomes to produce high-resolution mathematical diagnostics. The system enforces…
A developer has released an open-source supply-chain audit script, "Mini Shai-Hulud," designed to detect the TanStack Router and Mistral AI supply-chain attacks. The bash script checks for specific SHA hashes, malicious …
NVIDIA and SAP announced an expanded collaboration at SAP Sapphire to embed NVIDIA's open-source OpenShell runtime into the SAP Business AI Platform, providing security and governance controls for specialized AI agents i…
TanStack issued an all-clear on May 15, 2026, after a three-day security sweep following the compromise of 42 Router/Start packages on npm. The malicious code was injected through a poisoned CI cache, not through stolen …
A developer has created a structured prompt system for auditing Elixir and Erlang projects, defining two entry points: `audit_file/4` for single-file analysis and `audit_directory/2` for whole-package audits. The system …