AI Safety, Security, and Global Governance
Key Questions
Why are calls increasing for mandatory synthetic DNA screening?
AI systems now outperform PhD-level virologists, raising biosecurity concerns and prompting Anthropic and others to urge global pauses and mandatory screening protocols.
What recent policy actions affected frontier AI models?
The US government restricted foreign access to Anthropic's Fable 5 and Mythos 5, forcing suspension of new tools, while Anthropic released a policy framework for frontier AI risks.
How are vulnerabilities being addressed in MoE and VLM models?
Research shows MoE safety is concentrated in under 20% of experts, creating vulnerabilities, and Los Alamos PAS method detects VLM hallucinations via attention patterns at SOTA levels.
US government restricts foreign access to Anthropic's Fable 5 and Mythos 5, forces suspension of new tools over security concerns; open letter from researchers demands restoration. Miles Brundage criticizes Anthropic blog posts as effectively export-controlled. Pramaana Labs raises $27M for formal verification of AI. California releases model policy on AI in education. Claude Fable 5 introduces auto-routing guardrails and 30-day data retention. Anthropic apologizes for invisible guardrails; shadowbanning reported. Dario Amodei calls for mandatory third-party testing.