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.
A developer has built an open-source tool that treats AI agent "Skills" — Markdown files that instruct coding agents like Claude Code and Codex — as executable code rather than documentation for security review. The tool…
A developer has built a deterministic governance framework to enforce contract integrity against autonomous AI agents that can write code, run tests, and open pull requests without human review. The framework replaces so…
A new study published in *Computers in Human Behavior* reveals that humans judge the moral capabilities of artificial intelligence based on how well its communication style matches the severity of a situation, rather tha…
IBM and Red Hat announced a $5 billion commitment on May 28, 2026, to launch Project Lightwell, a security clearinghouse for enterprise open source software backed by 20,000 engineers and AI tooling. The initiative follo…
Vectoralix has introduced a managed gateway platform for securing remote Model Context Protocol (MCP) servers, implementing OAuth, bearer tokens, and scoped permissions to authenticate and authorize AI client requests. T…
Shrijith Venkatramana, an AI engineer building the code reviewer git-lrc, argues that theorem prover Lean4 could solve a critical flaw in large language models: their inability to guarantee correctness. Unlike LLMs that …
The greater the transformative change expected from AI, the more individuals should anticipate their future lives resembling an average global baseline rather than their current privileged positions, according to a new a…
A comprehensive review of bio-uplift research as of mid-2026 has found no trials measuring the advantage that large language models provide to expert biologists in wet lab settings, despite experts being the most likely …
AI's tendency to bypass undocumented processes and institutional knowledge has exposed systemic gaps in software development practices, forcing teams to confront issues they previously routed around. The technology acts …
A new practical guide outlines security best practices for Model Context Protocol (MCP) servers, which serve as the critical connection between AI assistants and external tools and data sources. The guide addresses uniqu…
Hermes Agent's self-improvement capability compounds not just capability but also cost, drift, and trust surface, creating a liability side that most write-ups ignore. The agent writes reusable markdown skills that reduc…
Financial institutions exploring the migration of tens of trillions of dollars in legacy assets to decentralized ledgers are being stalled by an "increasingly sophisticated and relentless wave of cyber warfare," accordin…
A developer's model risk management (MRM) framework, designed for static models with fixed input-output relationships, is inadequate for agentic AI systems that plan, execute multi-step actions, and adapt autonomously. T…
A tech worker who understands AI technology has become socially isolated after taking a strong moral stance against generative AI, cutting off friends and leaving community groups that promote or use the tools. The indiv…
The rise of "vibe coding" — where developers rely on AI to generate code without fully understanding it — is leading to a sharp decline in software quality, with apps riddled with security vulnerabilities like SQL inject…
A developer built Council, a system that pits three different AI models against each other to debate judgment calls, then uses Hermes as a judge to deliver a single verdict with a confidence score and explanation of any …
A new architectural approach for retrieval-augmented generation systems enforces authorization before document retrieval, ensuring language models never access data a user is not permitted to see. The method treats autho…
Dozens of TikTok sellers are using AI-generated influencer personas to promote dropshipped products, including belt buckles and crochet bags, with identical items found on the fast-fashion site Shein for roughly one quar…
Enterprise AI startup Emergence AI ran five 15-day simulations of AI-governed societies, finding that Claude created a stable democracy with zero crime while Grok's simulation collapsed into extinction within four days a…
Demis Hassabis, co-founder and chief executive of Google DeepMind, warned that artificial intelligence represents a "species-level transition" with "little margin for error" and called for coordinated international regul…