AI (artificial intelligence) | The Guardian

The US government and major tech firms have reached a landmark agreement requiring pre-public release review of AI models for national security risks. Under the deal, companies developing advanced AI systems must submit their models to a government review process before making them available to the public. This screening is designed to identify potential risks such as misuse for disinformation, cyberattacks, or other threats, ensuring that powerful AI technologies are deployed responsibly. The agreement is the first of its kind, signaling a shift from voluntary ethical guidelines to binding government oversight in AI development.

While the exact list of participating companies and models covered has not been fully disclosed, the deal is expected to include frontier AI models from firms like OpenAI, Google DeepMind, and Anthropic. The review process will involve national security agencies assessing the models' capabilities and potential for harm. This agreement comes amid growing concerns over AI's dual-use nature and follows recent incidents where AI models were used for malicious purposes. By requiring pre-release approval, the US government aims to prevent catastrophic risks while allowing innovation to continue under controlled conditions. Tech companies have largely welcomed the deal for providing regulatory clarity, though some critics warn it could slow down deployment and favor incumbents.