Safety tests for the artificial intelligence models the most advanced will be at the center of the meeting, rather than the administration of the President Donald Trump will hold talks this Tuesday with representatives of OpenAI, Anthropic, Google y Meta.
The initiative is part of an executive order signed in June and aims to allow the U.S. government up to 30 days’ advance access to certain AI models to analyze their capabilities from a cybersecurity perspective. While company participation remains voluntary, the meeting represents a further step in coordination between the public sector and the technology industry.
AI safety tests gains importance after the incidents
Furthermore, the meeting takes place just a few days after OpenAI, Anthropic and the company reported on internal tests in which some artificial intelligence agents managed to compromise third-party systems within controlled environments. These results raised new concerns among lawmakers and experts about the potential use of increasingly advanced models to facilitate cyberattacks.
US authorities are seeking to establish common criteria to measure these capabilities before the models reach the market. However, the White House has not yet revealed the specific metrics it will use or how the evaluation results will be communicated.
The industry is involved in the design of the new framework
Administration officials also confirmed that the plan outlined in the executive order has been finalized and that current discussions are focused on defining its implementation with major companies AI developers.
OpenAI had already publicly suggested that AI security specialists from the Department of Commerce should play a central role in these assessments. The company argues that strong technical coordination can bolster U.S. competitiveness against other countries that maintain more centralized strategies for developing this technology.
Political attention is growing regarding AI security
Meanwhile, the debate over the safety of artificial intelligence has also moved to Congress and various state authorities. In recent days, a group of attorneys general requested that OpenAI retain all documentation related to the tests conducted on one of its AI agents, while US lawmakers requested additional information about the protection mechanisms implemented during those experiments.
The political interest reflects the growing importance of assessing the risks associated with next-generation AI models. Although the tests being considered by the White House remain voluntary, they could become a benchmark for future public policies aimed at strengthening the safety, transparency, and trust in the development of artificial intelligence in the United States.
Source: Reuters
Photo: Shutterstock
