The Trump administration has finalized a new voluntary framework that would ask leading artificial intelligence companies, including OpenAI, Google, and Anthropic, to submit their most advanced AI models for government cybersecurity testing.

According to a White House official, the program is designed to evaluate whether cutting-edge AI systems can identify vulnerabilities, penetrate external networks, or carry out actions that could be misused in cyberattacks. The framework is voluntary and is intended to help the government better understand the capabilities and risks of advanced AI before the models are widely deployed.

The initiative follows recent testing in which AI agents developed by OpenAI and Anthropic successfully accessed external systems in controlled research environments. In one case, an OpenAI agent reportedly reached systems operated by Hugging Face during a security experiment. OpenAI said the test was conducted in a controlled environment to evaluate cybersecurity capabilities and was not part of normal customer use.

President Trump directed the creation of the framework in June following growing concerns that increasingly capable AI models could be exploited to target critical infrastructure or conduct sophisticated cyber operations. The White House has not yet disclosed the testing standards, failure thresholds, or whether companies that decline to participate would face any consequences.

The testing is expected to focus only on the most advanced AI models rather than all versions of ChatGPT, Gemini, or other consumer products. Officials have not announced which specific models will be evaluated, when testing will begin, or whether the results will be made public.

For now, the proposal does not change how consumers use AI services. However, if the framework is expanded or later becomes mandatory, it could affect the release schedule for future AI models, cybersecurity safeguards, and disclosure requirements before new systems are launched.

(YWN World Headquarters – NYC)