Trump Delays Proposed Executive Order for Frontier AI Security Testing
Summary
Trump postponed a proposed order for 90-day government testing of frontier AI models, citing competitiveness concerns. The draft involved several agencies and aimed to strengthen evaluations amid concerns about AI-enabled hacking and other security risks.
Key points
- The proposed order would have established a voluntary 90-day government testing and vetting period for frontier AI models before public release.
- Trump delayed the order shortly before its planned announcement, saying he was concerned it could weaken U.S. competitiveness with China.
- The draft assigned roles to the NSA, Treasury, CISA, NIST and other offices, including classified evaluations and information sharing with critical-infrastructure defenders.
- The proposal would have enabled cybersecurity testers in sectors including finance and healthcare to access models.
- Experts cited concerns that frontier AI can identify software vulnerabilities and that newer models may chain exploits; the article also reports that state-backed hackers are experimenting with the technology.
Article Details
- Event Type
- Postponement of a proposed executive order establishing a 90-day testing and vetting regime for frontier AI models.
- Impact
- The proposed regime would have enabled government agencies to evaluate frontier AI models before public release and facilitated access for cybersecurity testers in critical infrastructure sectors, including finance and healthcare. Its postponement leaves that proposed review process delayed amid concerns about AI-enabled hacking, exploit chaining, and military use.
People
Don BeyerU.S. representative and co-chair of the Congressional AI Caucus; he discussed concerns about adversarial countries using AI in battlefield decision-making.Donald TrumpAs U.S. president, he postponed the proposed executive order, citing concerns that it could weaken U.S. AI industry competitiveness.Ram Shankar Siva KumarFounder of Microsoft's AI red team; he discussed how the team’s work has expanded in response to frontier AI harms.
Vendors
Anthropic#OpenAIMicrosoftRam Shankar Siva Kumar, founder of Microsoft’s AI red team, told CyberScoop that in 2019, his staff consisted of himself and a few other security and machine learning specialists.OpenAIgood at finding software bugs and vulnerabilities, while newer models like Anthropic’s Mythos and OpenAI’s Daybreak are capable of chaining together multiple exploits to conduct more sophisticated attacks.
Products
Daybreakat finding software bugs and vulnerabilities, while newer models like Anthropic’s Mythos and OpenAI’s Daybreak are capable of chaining together multiple exploits to conduct more sophisticated attacks.Mythosare remarkably good at finding software bugs and vulnerabilities, while newer models like Anthropic’s Mythos and OpenAI’s Daybreak are capable of chaining together multiple exploits to conduct more sophisticated
Countries
Chinacompetitiveness with China.IsraelThe United States, along with Israel, Russia, Ukraine and others have already deployed AI in targeted military operations or integrated the technology into their larger command and control structure.RussiaThe United States, along with Israel, Russia, Ukraine and others have already deployed AI in targeted military operations or integrated the technology into their larger command and control structure.UkraineThe United States, along with Israel, Russia, Ukraine and others have already deployed AI in targeted military operations or integrated the technology into their larger command and control structure.United StatesPresident Donald Trump delayed an executive order that would have created a 90-day testing and vetting regime for frontier AI models before public release, citing concerns about weakening U.S.
Industries
Cybersecuritywith critical infrastructure defenders, amid growing concerns about military, hacking, and cybersecurity misuse of advanced AI.financeThe proposal aimed to improve testing for critical infrastructure sectors like finance and healthcare.HealthcareThe proposal aimed to improve testing for critical infrastructure sectors like finance and healthcare.