Trump administration expands AI model stress-test program to Google DeepMind, xAI, Microsoft
deepseek / deepseek-v4-flash
- The Trump administration expanded a program giving government scientists access to unreleased AI models for risk assessments.
- The program now includes Google DeepMind, xAI, and Microsoft, in addition to OpenAI and Anthropic.
- Scientists focus on ‘demonstrable risks’ such as cyberattacks and development of chemical or biological weapons.
- OpenAI is testing GPT-5.5-Cyber, a variant for defensive cybersecurity work.
- Anthropic revealed vulnerabilities that bypassed safety mechanisms, which have been patched.