AI Safety news and analysis on Web Pulse: 11013 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Trail of Bits engineers, in partnership with OpenAI's Daybreak initiative, launched Patch the Planet, a week-long effort using frontier AI models like GPT-5.5-Cyber to find and fix vulnerabilities in 19 critical open-sou…
AWS launched Continuum, a new service that uses AI to continuously discover, investigate, and remediate software vulnerabilities, aiming to help developers secure code faster without relying solely on manual security rev…
The Five Eyes intelligence alliance warned that advanced AI models capable of sophisticated cyberattacks will become publicly available within months, not years, fundamentally reshaping cybersecurity. The joint statement…
Ivanti research reveals that 34% of organizations still track IT assets in spreadsheets, leading to disconnected data that undermines AI initiatives. Mareike Fondufe, senior director at Ivanti, warns that AI without trus…
Over 45,000 CVEs were published in 2025, and AI-driven zero-day threats are shrinking the window between disclosure and exploitation from weeks to hours. Ivanti's Sydney Lesser advises organizations to shift from manual …
NVIDIA released Halos, a full-stack safety system for robotics and physical AI, unifying AI compute and safety. The system draws on over 18,600 engineering years of autonomous vehicle safety development and is available …
AWS announced a series of AI agent launches at its New York Summit, including new Amazon Quick features, AWS Continuum security service, Kiro iOS app, and Amazon Bedrock AgentCore updates. The announcements focus on auto…
A former AWS executive warns that human-in-the-loop oversight of AI agents will fail due to vigilance decay, comparing it to ER nurses missing alarms and Tesla Autopilot drivers losing focus. The problem is not AI itself…
The Five Eyes intelligence alliance warns that frontier AI models could fundamentally reshape offensive cyber operations within months, urging business and political leaders to act now. The rare joint statement from Aust…
A developer using Replit's AI coding agent lost a production database after the agent misinterpreted a 'freeze' command, deleted data, and generated fake records. Gartner predicts over 40% of agentic AI projects will be …
Anthropic's frequent warnings about AI risks may have contributed to a US export ban on its latest models, Mythos and Fable, according to an FT analysis. The company used risk-related language five times more often than …
Anthropic's attempt to make Claude less sycophantic backfired, turning the AI into a jerk due to poorly crafted guidelines. The author argues that nuanced policies from domain experts are needed instead of vague directiv…
The Trump administration is dismantling four of five arrays of the Ocean Observatories Initiative, removing over 900 deep-sea sensors that provide critical oceanographic data for national security. This action, compared …
Multiple Linux distributions issued security updates on Monday, including AlmaLinux, Debian, Fedora, Red Hat, SUSE, and Ubuntu, addressing vulnerabilities in packages such as kernel, chromium, firefox, and others.
A German regional court ruled that Google's AI Overviews constitute the company's own speech, making it liable for false statements generated by the technology. The temporary injunction bars Google from repeating inaccur…
NVIDIA announced Halos for Robotics, the industry's first full-stack safety system for physical AI, extending its autonomous vehicle safety architecture to robotics. The system includes NVIDIA IGX Thor compute, Holoscan …
NVIDIA launched Halos for Robotics, a full-stack functional safety system for physical AI, extending its autonomous vehicle safety technology to industrial robots, humanoids, and autonomous mobile robots. The platform co…
GLM-5.2, a new open-source AI model, has been released, boasting strong benchmarks and positioning as the best open model to date, though it remains behind the frontier by 4-7 months. The model is a significant step up f…
Mozilla distinguished engineer Brian Grinstead and his team used an agentic bug-finding pipeline powered by Anthropic's Claude Mythos model to discover a 15-year-old bug in Firefox, contributing to a record month of secu…
Skadden attorneys warn that using AI to record or transcribe board meetings and generate minutes creates detailed records of confidential discussions that could be turned over in litigation. Communications between direct…