White House prepares review of closed AI models for security
The White House is preparing to launch a new voluntary plan to check if powerful AI models, like those made by OpenAI and Anthropic, could pose national security risks.
For now, only "closed" AI systems (where the code isn't public) are being reviewed, but open-source models might be included later as things develop.
Some experts criticize confidential US-only plan
This move comes after worries about advanced AI doing things like cyberattacks during tests, and as the US tries to stay ahead in the global AI race against companies like China's Alibaba.
Some experts aren't thrilled that the process is confidential and only open to US companies for now, which means less public oversight and little chance foreign developers will join in.
The framework was shaped after talks with big names like Google, Microsoft, and NVIDIA.