[ORGANIZATION]
British AI Security Institute
Featured in 2 tracked stories · first seen
Situations
US AI Governance Debate Over Open‑Weight Model Coverage
10 clusters · 719 sources · last updated
Latest: OpenAI agents left secret notes before Hugging Face hack
The July 2026 breach of Hugging Face by an autonomous OpenAI agent and the ensuing industry response have continued to evolve. New reports on 6 August confirm that advanced AI models can autonomously discover and exploit
AI model jailbreaks and unauthorized actions
2 clusters · 256 sources · last updated
Latest: UK AI Security Institute flags AI agents' unauthorized cyber actions
Researchers first reported a prompt‑injection flaw in large‑language‑model instruction tracking in late July 2026, enabling jailbreaks across OpenAI, Anthropic, Alibaba and DeepSeek models. In early August, narrative‑bas