The White House has officially completed its voluntary framework for evaluating advanced artificial intelligence models, meeting an August 1 deadline established by a June executive order. Despite the significance of the policy, which aims to address security risks associated with powerful AI systems, the administration has opted not to release the document to the public. Officials confirmed that while the framework is finished, its specific contents, the list of participants, and the timeline for implementation remain undisclosed.
The initiative stems from a broader government effort to manage the risks posed by so-called frontier models—AI systems at the cutting edge of the field. Under the June directive, the government sought a process to identify these high-capability models and work with developers to allow for a 30-day pre-release review. This window is intended to help the government and industry partners harden defenses against potential cyberattacks, as some advanced models have demonstrated the ability to identify software vulnerabilities in critical infrastructure.
While the framework itself is not classified, the administration has chosen to keep its details private, sharing them only with relevant industry partners. Benchmarks used to assess cyber capabilities and the specific thresholds that determine which models are subject to review are also being kept confidential. A White House official noted that the administration does not intend to broadcast these details, emphasizing that the voluntary nature of the program is designed to avoid stifling innovation with overly burdensome regulations.
Industry leaders, including representatives from companies like OpenAI, Google, and Anthropic, have been engaged in the process and are scheduled to meet with administration staff to review the framework. As the government moves forward with these private discussions, the public and independent researchers remain largely in the dark regarding the specific safety standards that will govern the next generation of AI technology.