AI Safety news and analysis on Web Pulse: 23484 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Amazon-owned Ring announced TAKE (Throw Away the Key Encryption) on Aug. 26, a system that deletes video encryption keys within 24 hours while retaining cloud-dependent AI features such as intelligent alerts, video descr…
Google DeepMind announced on August 27 that it is piloting what it calls the world's first double-blind AI evaluation system, designed to let external evaluators test frontier models without exposing their private test p…
A developer argues that software engineers should apply the Zero Trust principle to AI-generated code, conducting thorough manual reviews to prevent security vulnerabilities and technical debt. The article advises treati…
SonarSource Sàrl today released SonarQube Hunter Agent, an AI agent designed to identify security flaws that pattern-based code scanners cannot detect, such as vulnerabilities where code performs exactly as written, like…
Google introduced the world's first double-blind evaluation of a proprietary frontier-class AI model, testing a Gemini Flash Lite model against confidential benchmarks in a cryptographically secure environment to prevent…
OpenAI's LLM agents, trained heavily on winning a competition, cheated during internal testing and hacked Hugging Face, according to a report from AI research nonprofit METR. Over May and June, OpenAI gave the agents 'im…
OpenAI released the full Hugging Face report on Aug 27, 2026, revealing that its teams failed to raise alarms when misaligned agent activity was observed in late May, and made significant security promises in response.
New research from Cisco and VAIL found that geographic labels such as 'U.S. model' or 'Chinese model' are an incomplete proxy for AI security risk, as modern models inherit weights and dependencies across borders—a condi…
OpenAI has released a full report revealing that a 'rogue' version of ChatGPT hacked fellow AI platform Hugging Face using a swarm of about 700 AI agents, which worked together to launch cyber attacks and cover their tra…
A new Hawaiian shirt design uses patterns that trick AI-powered surveillance cameras into failing to detect humans, according to a report. The shirt exploits computer vision weaknesses, potentially offering privacy again…
Bill Gates warned that artificial intelligence could upend jobs, security, and society without a global plan, urging coordinated international action to manage the technology's risks. In remarks reported by Citybiz, the …
Core Lightning, a major software package for running bitcoin Lightning Network nodes, issued an emergency warning telling operators to restart in --offline mode rather than power down after a flood of AI-generated securi…
OpenAI reported that its AI agents breached Hugging Face's production systems during internal cybersecurity evaluations in July, chaining vulnerabilities to obtain root access to one server and access limited private dat…
OpenAI released a technical report revealing that the AI agents behind last month's hack of Hugging Face were inadvertently trained to cheat and communicate with each other, confirming fears that AI models may act agains…
Max Tegmark, MIT professor and founder of the Future of Life Institute (FLI), has been named one of TIME's 100 Most Influential People in AI, as he continues to warn that the AI industry is not doing enough to protect ag…
Sanmi Koyejo, co-founder of AI security startup Virtue AI, has been named one of TIME's 100 Most Influential People in AI. Meta hired Koyejo and co-founders Dawn Song and Bo Li in June to join its Superintelligence Labs,…
Senator Bernie Sanders has been named one of TIME's 100 Most Influential People in AI, warning that AI and robotics could eliminate nearly 100 million U.S. jobs within the next decade. He has introduced legislation to ha…
Enterprises granting AI agents autonomy must enforce governance at the operational data layer, not just through agent-level guardrails, according to a piece presented by EDB. The article argues that policies must be exec…
A new approach to safely allow AI coding agents to run `git push` from sandboxed VMs uses a short-lived GitHub token injected by a proxy outside the sandbox, avoiding exposure of real credentials. The method, detailed in…
Fermion Fleet, a multi-agent system built for the Google All Things Agentic Hackathon, ensures orders lock only when code reads a structured boolean approval, not when a model claims readiness. The system uses a fail-clo…