AI Will Be Controlled Before Reaching the Public: US Government Agreement with Google DeepMind, Microsoft, and xAI

According to reports from the past few hours, the US government has announced that it will strengthen its role in overseeing advanced artificial intelligence models by forming new agreements with some of the major players in the industry. The National Institute of Standards and Technology, through the Center for AI Standards and Innovation (CAISI), has announced a review of its collaborations with Google DeepMind, Microsoft, and xAI to evaluate cutting-edge models before their public release.

This initiative is part of a broader strategy for monitoring and understanding the capabilities of advanced AI, with a specific focus on impacts in the area of national security. The agreements indeed allow the government to access models in the pre-deployment phase, as well as conducting analyses after release, with targeted research activities.

According to what has been communicated by CAISI, the companies involved will allow the analysis of their systems before commercial distribution. In some cases, the models provided to evaluators may have reduced or removed protections, allowing for a deeper analysis of capabilities and potential risks. This operational mode aims to provide a more realistic snapshot of AI system capabilities, going beyond the limitations imposed by public versions.

Evaluations are conducted with contributions from experts from various government agencies, coordinated through task forces dedicated to national security. To date, the center has already completed over 40 evaluations of advanced models, including systems not yet publicly released. Current collaborations represent an extension of already existing partnerships, updated to align with new federal directives regarding AI. Activities also include sharing information with companies, aiming to encourage voluntary product improvements.

CAISI has been designated as the primary reference point for the AI industry within the US government. Its role includes facilitating testing, promoting joint research, and developing best practices for commercial AI systems. The new agreements could represent a step toward a more structured governance system for artificial intelligence in the United States. According to some reports, the administration is considering further measures, including the creation of direct coordination between technology companies and institutions for supervising advanced models.