Trump Signs Executive Order For AI Testing Prior To Frontier Model Releases

2023-08-29 · Source: Don't Worry About the Vase · Field: Government & Public Sector — Public Policy & Governance, Regulatory & Compliance · Depth: Intermediate, long

Summary

President Trump signed an Executive Order (EO) mandating pre-release testing for "frontier" AI models, reversing an earlier cancellation. The EO establishes a "voluntary" framework requiring labs to submit models for classified benchmarking and provide the government up to 30 days of confidential access before release to "trusted partners." This period was reduced from an initial 90 days. Despite the "voluntary" designation, the order is widely interpreted as a de facto mandatory licensing and prior restraint regime, particularly for companies seeking federal contracts. Key concerns include the central role of the NSA over civilian agencies like CAISI, the classified nature of benchmarking rules, potential for the government to pick "winners and losers" through early access control, and the EO's limited scope to only cyber threats, ignoring other catastrophic risks.

Key takeaway

For Directors of AI/ML developing frontier models, this Executive Order signals a new era of de facto mandatory pre-release government oversight. You should prepare for submitting models for confidential evaluation and understand that non-participation could impact federal contracting opportunities. Be vigilant regarding the classified nature of testing protocols and the NSA's expanded role, as these elements introduce potential for regulatory overreach and influence on market access.

Key insights

Trump's AI Executive Order establishes a de facto mandatory pre-release testing regime for frontier models, despite being framed as voluntary.

Principles

• AI regulation is an inevitable policy direction.
• "Voluntary" frameworks can become de facto mandatory.
• Classified regulatory processes introduce risks of abuse.

Method

The EO coordinates agency heads to define "covered frontier models" via classified cyber benchmarking, then offers a voluntary framework for labs to submit models for up to 30 days of confidential government access before broader release.

In practice

• Prepare for pre-release model evaluations.
• Anticipate NSA involvement in testing.
• Monitor classified benchmarking criteria.

Topics

• AI Regulation
• Executive Order
• Frontier AI Models
• Cybersecurity Testing
• National Security Agency
• Model Pre-clearance

Best for: CTO, Executive, VP of Engineering/Data, Policy Maker, Legal Professional, Director of AI/ML

Related on AIssential

• Trump Signs AI Executive Order to Increase Government Oversight — The Trump administration is tightening AI oversight, requiring 30-day pre-release access to powerful models and addressing software vulnerabilities.
• Trump plan to test AI models has a problem—US security teams were gutted by DOGE — Trump's voluntary AI safety testing EO faces significant implementation challenges due to limited government capacity and reliance on industry goodwill.
• Trump abruptly cancels EO signing event after top AI firm CEOs declined to go — The US faces internal and external pressures balancing AI innovation with government-mandated safety testing, impacting its global regulatory standing.
• Trump signs executive order seeking early access to new AI releases — US AI policy balances national security concerns with a strong preference for innovation over mandatory regulation.
• Trump delays AI security executive order: ‘I don’t want to get in the way of that leading’ — Trump delayed an AI security executive order, fearing it would impede US leadership and due to a requirement for pre-release model sharing with the government.
• Read Trump's unsigned AI executive order - Politico — A proposed U.S. executive order aimed for voluntary AI oversight, balancing innovation with national security concerns.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Don't Worry About the Vase.