- President Trump abruptly postponed signing an executive order on AI cybersecurity Thursday, saying he “didn’t like certain aspects of it” and worried it could act as a blocker to US technological dominance over China.
- The order would have revamped cybersecurity information-sharing programs to include AI companies and called for voluntary government testing of frontier AI systems — stopping well short of mandatory federal approval of cutting-edge models.
- The last-minute reversal throws the administration’s AI policy agenda into uncertainty at a moment when Anthropic’s Mythos model has raised acute national security concerns for its ability to find network vulnerabilities at scale.
- Google, Microsoft, and xAI have already agreed to give the government access to their models for security testing; OpenAI and Anthropic were already part of an existing voluntary program through the Commerce Department.
What Happened?
With invitations already sent to technology executives for a White House signing ceremony, President Trump pulled the plug on an AI cybersecurity executive order hours before it was set to be signed. Speaking at the White House, Trump said he postponed it because he objected to parts of the directive and did not want any measure that could impede the US lead over China in artificial intelligence. The order had been weeks in the making, involving White House Chief of Staff Susie Wiles, Treasury Secretary Scott Bessent, National Cyber Director Sean Cairncross, and science adviser Michael Kratsios. It would have avoided mandatory oversight of AI models in favor of voluntary government testing programs to identify and patch security weaknesses in federal, state, and critical infrastructure systems.
Why It Matters?
The last-minute reversal highlights the fundamental tension running through Washington’s AI policy debate: how to address genuine national security risks from powerful AI systems without erecting guardrails that slow down US development relative to China. That tension has been made concrete by Anthropic’s Mythos model, which the company disclosed last month is extraordinarily effective at finding network vulnerabilities — a capability the NSA is already exploiting for defense but that also represents a major offensive risk. The White House has already blocked Anthropic from distributing Mythos more broadly, citing security concerns. Without an executive framework in place, ad hoc decisions like that one will continue to govern how frontier AI capabilities are managed across government.
What’s Next?
The White House will need to rework the order to address whatever specific provisions Trump objected to before attempting another signing. OpenAI has separately announced a partnership with the administration on a deployment strategy for GPT-5.5-Cyber, a model designed for cyber defense — suggesting the private sector is moving ahead with security applications regardless of the executive order’s fate. Watch for whether the administration attempts to replace the broad EO framework with narrower, model-specific agreements, and whether Congress steps in to fill the policy vacuum with its own AI security legislation.
Source: Bloomberg
