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.
Researchers with the Northern Tornadoes Project identified two widely shared images of tornadoes near London, Ontario as AI-generated fakes, with tornadic features digitally added to real photos. The NTP confirmed the Ma…
Detectify launched the Detectify MCP (Model Context Protocol) Server, an integration layer that embeds its security testing engines into AI-driven development workflows. The server exposes scanning capabilities to AI age…
A federal appellate court flagged an attorney's brief for containing fabricated case citations and fictitious quotations, the "hallmarks" of generative AI misuse, in a defamation lawsuit against members of an "Are We Dat…
Anthropic and its Project Glasswing partners have identified more than 10,000 high- or critical-severity vulnerabilities in critical software systems. The company introduced Claude Mythos Preview in April 2026, a large l…
The PUBLIC LIGHTHOUSE CORE v1.7 runtime has been released, adding two new sublayers to Layer 3 (State Validation): the Orientation State Register (OSR) for spatial frame and basis tracking, and Narrative State Validation…
Detectify launched the Detectify MCP Server, a new integration layer that embeds its security testing engines directly into AI-driven development workflows to help coding agents identify and validate exploitable vulnerab…
A model upgrade that alters escalation behavior, refusal patterns, tool posture, latency, or cost constitutes a production release, not a settings change, and must be subject to release gates, regression testing, runtime…
Elder Gerrit W. Gong of The Church of Jesus Christ of Latter-day Saints delivered a keynote address at the AI Summit on Ethics and Artificial Intelligence in Athens, Greece, calling for AI systems to be programmed with a…
In 2026, adversaries are integrating artificial intelligence across the entire cyber attack lifecycle, using automated reconnaissance, AI-generated phishing, and adaptive malware to scale and customize operations at unpr…
A solo developer built Loopers, an open-source reverse proxy that enforces hard dollar caps on AI API spending, after discovering developers were facing bills as high as $80,000 from runaway LLM costs. The tool intercept…
Malicious actors are increasingly using artificial intelligence to automate reconnaissance and exploit system vulnerabilities, making Distributed Denial of Service (DDoS) attacks faster, stronger, and harder to stop. A w…
Anthropic introduced 28 integrations between Claude and third-party security and compliance platforms on May 21, 2026, via the Claude Compliance API. The integrations give enterprise teams programmatic access to conversa…
Anthropic and OpenAI publicly clashed with the U.S. Pentagon over military AI use, issuing statements in late February 2026 that banned mass domestic surveillance and certain autonomous lethal functions from their govern…
AI penetration-testing startup Novee launched Agentic Fix, a capability that routes validated exploit findings and remediation guidance directly into developer coding assistants such as Claude, Codex, Copilot, Cursor and…
The European Central Bank convened lenders to address cybersecurity flaws exposed by AI models and pressed them to accelerate IT system remediation. The ECB-organized gathering focused on AI-exposed vulnerabilities and u…
Check Point has been running a Frontier AI Models Readiness Program to prepare its products for attacks enabled by advanced generative models, including large-scale AI-driven code scanning and accelerated protection deve…
Rogue states are deploying AI agents to automate sanctions evasion at industrial scale, according to a new report from the Royal United Services Institute (RUSI). The think tank warned that adversaries are using artifici…
A developer argues that infrastructure drift is inevitable and that teams should shift from prevention-focused to detection-first architectures. The analysis identifies three distinct drift origins—human, system, and pro…
A 21-year CTO and Ph.D. in computer science released three of six planned episodes in an educational AI video series, paired with a Creative Commons-licensed GitHub repository of scripts and artifacts. The project, built…
Hardware restrictions proposed to prevent dangerous superintelligent AI, such as limiting fabrication to 28 nanometer process nodes or computing capacity below 15,840 TFLOP/s, may inadvertently guarantee the creation of …