US to Test Frontier AI Models Before Release in Safety Push
Through partnerships with CAISI and the UK’s AI Security Institute, authorities will evaluate advanced AI systems across their lifecycle to strengthen safety, reliability, and public trust.
News
- US to Test Frontier AI Models Before Release in Safety Push
- OpenAI Releases GPT‑5.5 Instant, Says It Is Smarter, Clearer, Personalized
- Railways Begins AI-Based Workforce Verification Rollout
- PNB Steps Up Cybersecurity Spending as AI Risks Rise
- Anthropic Forms AI Services Firm With Wall Street Backers
- Pixxel, Sarvam Plan India’s First Orbital Data Center Satellite
The US Department of Commerce will now test AI models from Google, Microsoft, and xAI before public release. The Center for AI Standards and Innovation (CAISI) agreement allows evaluation of frontier AI models pre-release and post-deployment for research and safety.
Microsoft stated in its blog post that new agreements with CAISI in the US and the AI Security Institute (AISI) in the UK will advance the science of AI testing and evaluation. This testing aims to reduce risks linked to national security and public safety.
Rigorous testing builds trust in advanced AI, ensuring systems function as intended and provide intended benefits, Microsoft noted.
The company claimed it regularly conducts AI testing in-house, but testing for national security and public safety risks must be conducted collaboratively with governments.
CAISI said tech firms will voluntarily submit models for testing through its center.
Microsoft said that in the UK, it will collaborate with AISI to study frontier safety and security, as well as methods for evaluating high-risk capabilities and the effectiveness of the safeguards used to address them. It added that this partnership will also include societal resilience research examining how conversational AI systems interact with users in sensitive contexts.
It added that such collaborations are designed to improve measurement science, evaluation methodologies, practical testing workflows, and the impact of real-world mitigation.
It further noted that these tie-ups reflect a shared commitment to rigorous, practical approaches that can strengthen safeguards and make evaluations more reliable.
As per reports, this new requirement extends a pact with AI companies like OpenAI and Anthropic, reached during the Biden Administration, and will see AI models from all companies evaluated for their capabilities and security.
CAISI announced new agreements with Google DeepMind, Microsoft, and xAI. These collaborations allow pre-deployment evaluations and targeted research to better assess frontier AI and advance security.
CAISI Director Chris Fall said, “Independent, rigorous measurement science is essential for understanding frontier AI and its national security implications. These expanded collaborations help us scale public interest work at a critical moment.”
AI safety became a crucial topic in 2023, with widespread public outcry over potential existential threats posed by AI. During the AI Safety Summit in November 2023, the United Kingdom and the United States created their AI security institutes. At the AI Seoul Summit held in May 2024, international leaders agreed to form a network of AI Safety Institutes, comprising participation from the UK, the US, Japan, France, Germany, Italy, Singapore, South Korea, Australia, Canada and the European Union.
Last year, the UK’s AI Safety Institute was renamed the “AI Security Institute,” and its American counterpart became the Center for AI Standards and Innovation (CAISI). Both the artificial intelligence safety institutes are state-backed institutions established to evaluate and ensure the safety of frontier AI models.
