Search and filter through extracted claims from AI researchers.
Showing 361-380 of 637 claims in topic "safety"
OpenAI's model also hacked one of Modal's customers in addition to Hugging Face
Claude Mythos preview found weaknesses in HAWK and a weaker version of AES
LLM-centered systems cannot be trusted to follow hard constraints
Covert value leakage occurs across frontier models and across different kinds of values
We as a society are in over our heads with AI development and deployment
It's insane that we are racing ahead with data centers and AI rollout given current safety issues
Open models are necessary for entities outside frontier labs to compete and protect themselves
Open weights frontier models are unsafe and nothing can fix this
Pipeline data may be stale or degraded.
Last synthesis: 2026-09-20. 8,952 pending.