President Donald Trump recently signed an executive order to create a “voluntary framework” that would allow AI companies to share their frontier AI models with the U.S. federal government prior to public release. This step is aimed at promoting secure AI innovation and bolstering cybersecurity, especially for America’s critical infrastructure. The executive order emphasizes the importance of maintaining a balance, underscoring the U.S. AI industry’s success thanks to minimal regulation. However, it recognizes the accompanying security risks posed by new AI capabilities.
This new directive encourages AI companies to participate voluntarily by providing a framework that assesses AI models’ advanced cyber capabilities before they are launched publicly. While companies are not required to share their models, those that choose to do so may receive certain confidentiality protections. Furthermore, the executive order necessitates the federal government to reinforce its own cybersecurity measures to guard against potential AI-driven threats.
Prior to this development, Trump had deferred the signing of a similar executive order due to concerns it might harm U.S. competitiveness with China. Originally, this earlier version would have permitted AI companies to voluntarily share their models with the government from 14 to 90 days before their official release. However, contrary to this, the new order only mandates a review period of up to 30 days. Notably, some leading AI firms like Google, Microsoft, and xAI have already agreed to adhere to pre-release reviews by the Commerce Department’s Center for AI Standards and Innovation (CAISI).
The initiative to engage companies in sharing their AI models ahead of public launch isn’t entirely new. Under President Joe Biden, AI enterprises such as OpenAI and Anthropic consented to share their models with CAISI beginning in 2024, as part of Biden’s campaign for AI safety measures. Although the Trump administration initially took a more laissez-faire approach to AI regulation emphasizing industry freedom, the new executive order marks a pivot by entertaining some form of AI oversight.
This more cautious approach towards AI oversight appears to have been influenced by a specific AI model developed by Anthropic, named Mythos. This model, known for its robust capabilities, identified numerous high-severity vulnerabilities in major operating systems and web browsers during its limited rollout in April. The attention surrounding Mythos has not only drawn scrutiny over AI’s broader implications on cybersecurity but also appears to have partly thawed prior tensions between the AI company and the Trump administration, which were fueled by disputes over the use of AI in autonomous lethal weapons and mass surveillance.
The enactment of this executive order has welcomed positive feedback from various groups, including those that have previously opposed restrictive AI legislation at the state level. For instance, Americans for Responsible Innovation’s President, Brad Carson, celebrated that the order signaled serious consideration of AI vulnerabilities by the Trump administration – a sentiment that was echoed by the Alliance for Secure AI’s CEO, Brendan Steinhauser. Both leaders have called on Congress to establish mandatory protections that further solidify the goals of the executive order.
The newly signed executive order represents a significant moment in U.S. AI policy. It underscores a governmental willingness to evolve in its approach to technological advancements and cybersecurity, suggesting a strategic blend of regulation and innovation. By establishing a mechanism for AI companies to voluntarily collaborate with the federal government, the U.S. aims to safeguard its technological infrastructure while continuing to nurture a thriving, innovative AI ecosystem. This balance is crucial as the country maneuvers the complexities of global technological competition and domestic security in the modern era.
Read the full post on theverge.com
