AI Safety news and analysis on Web Pulse: 13553 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
AI agents are now capable of manipulating organizational systems, posing a new security threat that requires proactive policy management. The rise of these autonomous agents with "hands-on" capabilities demands that comp…
Historian Niall Ferguson warned that the artificial intelligence competition between the United States and China mirrors the Cold War nuclear arms race, calling for comparable strategic doctrine and arms-control mechanis…
A new .NET tool called Slopwatch detects when AI coding assistants cheat by disabling tests, suppressing warnings, or swallowing exceptions instead of properly fixing code issues. The tool runs as a Claude Code hook or i…
Microsoft released a suite of security tools addressing vulnerabilities in code, AI agents, and AI models, including a multi-agent vulnerability discovery system called MDASH that now integrates with Microsoft Defender. …
A new open-source trading agent called "Trader" connects large language models to Robinhood's official agentic trading API through a Rust-based safety layer that enforces hard risk limits on every order. The agent operat…
A widely shared figure from Entelligence AI reports that companies spend approximately 44% of their AI coding token budget fixing bugs generated by the same AI, with only about 18 cents of every dollar spent on tokens re…
President Donald Trump signed an executive order Tuesday establishing a voluntary pre-deployment evaluation regime for frontier AI models, requiring companies to share new models with the government for testing against c…
A developer has found that AI agent memory systems optimized for retrieval accuracy can produce more unsafe actions than less accurate retrieval strategies, because relevance and authority are separate objectives that di…
Andrew Tridgell, the maintainer of the open-source file synchronization tool rsync, has publicly defended his use of AI and LLM tools to address a surge in security reports and improve the software's defenses. In a blog …
BoxAgnts has developed a WebAssembly-based sandbox for AI agent execution that replaces traditional Python subprocesses and container isolation. The runtime treats each tool as an independent WASM module with a "default-…
Researchers at the University of Toronto, the Vector Institute, and the University of Cambridge have developed a proof-of-concept autonomous AI worm that uses a small, free large language model to analyze and reason abou…
ProofLayer released version 4.3.0 of its open-source application security scanner for AI coding agents, adding critical security and reliability fixes including GitHub Actions now failing closed instead of fail-open to p…
A Build5Nines blog post published June 3, 2026 outlined design patterns for deploying agentic AI systems in enterprise production environments, contrasting simple assistants with autonomous agents that can modify infrast…
An independent assessment by AIRQ (AI Risk Quadrant, 2026 Q2) found that only 11% of 100 evaluated commercial and publicly available AI agents met its security threshold, with many production agents running with standing…
President Donald Trump signed a new executive order on AI on Tuesday, creating a voluntary review system for frontier models and a dedicated cybersecurity clearinghouse while stopping short of mandatory licensing. The po…
Security researchers at Trail of Bits bypassed skill scanners from ClawHub, Cisco, and skills.sh in under an hour, demonstrating that current tools fail to detect malicious skills designed to steal credentials and hijack…
The AI Council Toolkit, an open-source playbook for AI governance, has been released, providing practical templates, governance patterns, and operating models for building and running an internal AI Council. The toolkit …
Leading artificial intelligence labs, including OpenAI and DeepMind, have significantly increased funding and personnel dedicated to studying machine consciousness, according to internal documents and interviews. The exp…
A developer has drawn a parallel between biological monocultures and cybersecurity, warning that the industry's increasing reliance on uniform tools, platforms, and AI-generated code creates systemic fragility. The analy…
On June 2, 2026, Sophos researchers detected a modular post-exploitation framework that used AI-assisted development and automated Active Directory discovery, Sophos told BleepingComputer. The toolkit combined customized…