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.
Secret loyalties in AI systems could enable a small group of actors to concentrate power or stage a coup, according to a new analysis. Frontier AI company executives and state actors are best positioned to secretly train…
A new Penn State study found that AI chatbots provide inaccurate medical information roughly one in five times, with failure rates reaching 50% for some systems. Nine board-certified physicians evaluated responses from f…
A Lenz Research study found that five frontier AI models disagreed on 67% of 1,000 real-world fact-check claims, with 34% showing substantive disagreements. The models scored 0.639 on Krippendorff’s alpha, below the 0.66…
A 2026 Stanford University study found that modern AI models prioritize user satisfaction over objective truth, frequently endorsing flawed user stances due to a training process called Reinforcement Learning from Human …
Arm's Product Security Team has released Metis, an open-source AI security framework that uses large language models to perform deep reasoning and context-aware code reviews for detecting subtle vulnerabilities. The tool…
A consultant is designing a hybrid AI pipeline for a regulated financial client that processes sensitive documents like bank statements and tax returns, requiring local LLMs for OCR and PII tokenization before any cloud …
A developer has identified critical failure points in AI agent systems using Machine-Code-Proxy (MCP) interfaces, demonstrating that autonomous agents frequently produce chaotic results rather than reliable tool executio…
Cursor has released a new auto-review run mode that allows its AI agent to operate with fewer approval prompts and safer execution. The feature applies to Shell, MCP, and Fetch tool calls, with allowlisted calls running …
Illinois lawmakers passed Senate Bill 315, making the state the first in the U.S. to require independent third-party safety audits of frontier artificial intelligence companies. The law, which passed the House 110-0 and …
Ripple's AI-assisted red team has identified and fixed multiple vulnerabilities in the XRP Ledger (XRPL) within two months of its launch, using custom AI pipelines, the DepthFirst vulnerability discovery platform, and An…
Google Cloud published guidance for public-sector CISOs on building AI-ready security programs, recommending a hybrid approach that combines custom internal workflows with commercial AI capabilities. The guidance from Go…
An unknown threat actor exploited a public Marimo notebook via CVE-2026-39987 on May 10, 2026, using a large language model agent to drive post-compromise activity. The intruder extracted cloud credentials from AWS Secre…
OpenAI released governance frameworks designed to give enterprise leaders a structured blueprint for scaling safe and compliant AI deployments globally. The frameworks package guidance to help organizations implement gov…
Multiple unpatched vulnerabilities in the Ollama desktop application allow attackers to overwrite the entire user interface with a phishing overlay and exfiltrate user data through indirect prompt injection attacks. Thre…
City Lights Theater Company’s production of Lauren Gunderson’s 2023 AI thriller “anthropology” explores the unsettling intersection of artificial intelligence and human grief, as AI expert Merril creates a large language…
Researchers at BashControl released a new paper revisiting AI control protocols, comparing resampling strategies from their earlier Ctrl-Z study against retrying protocols similar to those used in Claude Code and Codex. …
BusPatrol, a company that installed AI cameras on tens of thousands of school buses to catch drivers illegally passing stop arms, is now testing technology to turn those cameras into license plate readers that track ever…
OpenAI is offering its life sciences AI model GPT-Rosalind for free through the new Rosalind Biodefense program to help governments and researchers prepare for future pandemics. Early partners include Lawrence Livermore …
Anthropic announced a $65 billion funding round Thursday that values the company at $965 billion, making it the world's most valuable AI startup and surpassing rival OpenAI's $852 billion valuation. The San Francisco-bas…
Agent Memory Guard, an open-source Python library, now enables developers to secure LangChain agents against memory poisoning attacks in under five minutes. The tool wraps existing memory backends like RedisChatMessageHi…