AI Safety news and analysis on Web Pulse: 13139 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Amazon CEO Andy Jassy reportedly raised security concerns about Anthropic's AI models with White House officials, leading the U.S. government to issue an export control directive that forced Anthropic to disable its Fabl…
Three vulnerabilities in LangGraph's checkpointer, disclosed on June 12, allow attackers to chain SQL injection into remote code execution on self-hosted servers using SQLite or Redis backends. The flaws, discovered by C…
Leaders of the Group of Seven (G7) nations are convening in Évian-les-Bains, France, from June 15 to 17 for the 52nd G7 Summit, focused on global economic stability, international security, and the future of AI. French P…
Anthropic's 2026 Agentic Coding Trends Report reveals that developers use AI for roughly 60% of their work but can fully delegate only 0–20% of tasks, creating a 40-point gap that represents the core enterprise challenge…
A new study reveals that AI-generated sexualized imagery and propaganda targeting Muslim women in India is on the rise, with over 6.7 million interactions on social media platforms. Victims like Samreen Ayoub report deep…
General-purpose large language models outperformed specialized clinical AI tools on all three medical benchmarks, according to a study published in Nature Medicine. The findings highlight the need for independent, real-w…
Broadcom released the largest set of Spring security updates in the framework's history on June 8, 2026, prompted by a 1,700% jump in CVE reports driven by AI-assisted vulnerability scanning. The most critical flaw, CVE-…
Researchers introduced VFUSE, a mechanistic interpretability approach using sparse autoencoders to audit protein models for hazardous features. Applied to RoseTTAFold3 and RFDiffusion3, linear probes in SAE latent space …
A developer on Qiita built a free CLI tool that runs a 30-second security scan on AI-generated code, but an engineer warns that such scans can create a false sense of safety. The engineer recounts a production incident w…
Anthropic released Claude Fable 5, its most capable model for long-horizon agentic work, on June 9, 2026, but a U.S. government export directive forced it offline within three days. The model shares the same underlying a…
Researchers at the University of Tennessee and the University of Florida developed a hardware neural network backdoor that can be hidden in custom chips used for edge AI, exploiting vulnerabilities in the supply chain of…
BioMedixAI, an AI-native healthcare platform, underwent a full-day adversarial audit to ensure production readiness. The team aligned vital-sign thresholds with the NEWS2 standard, fixed timezone issues by using local mi…
A California appellate panel upheld $6,000 in sanctions against a law firm for submitting a brief containing fictitious case citations generated by AI, which a trial judge described as the worst misconduct he had ever se…
Defense contractors including Anduril, Palantir, and Lockheed Martin are partnering with frontier AI companies to integrate advanced models into military systems that automate drone tasking and kill-chain proposals. The …
An anonymous user on Hacker News asked how Anthropic obtained private security bug data for training its Mythos or Fable AI models, noting that Anthropic's documentation only mentions publicly accessible data. The user s…
A new AI agent pre-execution checklist proposes that humans declare intent and boundaries in natural language, which AI converts into executable JSON to operate autonomously but stop when uncertain, enabling safer physic…
A coalition of state attorneys general has launched an investigation into OpenAI, the maker of ChatGPT, over potential consumer protection violations, according to a Wall Street Journal report. The probe focuses on wheth…
Senior engineers at U.S. technology companies are spending their weeks cleaning up AI-generated code, which, despite receiving high quality ratings during review, leads to more production incidents once deployed. The shi…
Researchers found that repetition loops in Gemma 4 instruction-tuned LLMs can be fixed by editing as few as one neuron, but the fix does not resolve deeper 'doom loops' caused by missing knowledge. The study shows that w…
Researchers propose RT-VLA, a lightweight vision-language-action model for autonomous driving that uses knowledge distillation from the larger SimLingo model to achieve real-time performance. RT-VLA reduces inference lat…