Key Takeaways
- President Trump signed an executive order that creates a voluntary process for top AI firms to share upcoming models with the government for safety testing up to one month before release.
- The order largely formalizes existing informal agreements; it does not impose new mandatory requirements on companies.
- Despite proclaiming a “common‑sense” balance of innovation and security, the administration’s AI policy has been marked by frequent reversals and contradictory actions.
- Former advisers and outside commentators disagree on whether the order represents meaningful progress or a watered‑down gesture.
- The White House simultaneously weakened the Cybersecurity and Infrastructure Security Agency (CISA) while tasking it with implementing the order, highlighting internal inconsistencies.
- Relations with Anthropic illustrate the administration’s mixed signals: praising the firm’s leadership while legally blocking its national‑security work and restricting access to its advanced models.
- Trump’s rhetoric about beating China in AI coexists with policies that allow advanced chips to flow to Chinese firms and with voluntary pledges to limit data‑center electricity costs.
- Overall, the executive order serves more as a political signal than substantive regulation, leaving a narrowing window for effective government oversight of rapidly advancing AI technology.
Overview of Trump’s AI Policy Shift
For months the White House hinted at a possible crackdown on the AI industry, only to reverse course repeatedly. In early 2025 Trump rescinded several modest Biden‑era AI safeguards, labeling them “dangerous” barriers to American leadership. Yet just weeks later he announced a new executive order that, while framed as a proactive step, largely mirrors the very voluntary mechanisms he had previously dismissed. This pattern of announcing a tough stance, then pulling back, then announcing again reflects a broader inconsistency that has characterized his approach to AI since returning to office.
Details of the New Executive Order
The order signed today establishes a voluntary process whereby leading AI companies—such as OpenAI and Anthropic—may share forthcoming models with federal agencies for safety testing up to one month before public release. It also asks firms to collaborate on bolstering federal, state, and local cyberdefenses. White House spokesperson Liz Huston described the policy as a “common‑sense approach of collaborating with industry to balance innovation and security.” Importantly, the order does not create new enforcement mechanisms or penalties; it merely formalizes information‑sharing arrangements that were already in place through ad‑hoc agreements.
Continuity with Biden‑Era Policies
Analysts note that the substance of today’s order closely resembles provisions from the Biden administration that Trump had previously condemned. Daniel Remler of the Center for a New American Security observed that the order “effectively formalizes what has already been happening between the US government and the leading AI companies.” Similarly, Dean Ball, a former Trump AI adviser, argued that the policy is “considerably more intrusive” than Biden’s executive order, suggesting that the administration’s rhetoric of deregulation belies a regulatory framework that is, in many respects, unchanged—or even expanded—relative to its predecessor.
Reactions from Former Advisers and Commentators
The order has elicited mixed responses from figures who once shaped Trump’s AI policy. Dean Ball expressed skepticism, viewing the measure as overly prescriptive. In contrast, David Sacks, the former White House AI czar who had originally opposed a more burdensome version of the order, later praised the final text as a “game changer” on social media, despite acknowledging that the government‑review process is little different from what he had previously rejected. Adding another layer, Steve Bannon—a former chief strategist and vocal AI critic on the right—called the order “directionally… pretty damn good,” arguing that codifying any rules, even weak ones, moves the debate forward.
Contradictions Involving CISA and Cybersecurity
One striking inconsistency lies in the administration’s treatment of the Cybersecurity and Infrastructure Security Agency (CISA). While the executive order tasks CISA with helping implement the voluntary AI‑testing framework, the Trump administration has simultaneously moved to gut the agency’s budget and staffing levels, undermining its core mission of protecting the nation from cyber threats. This juxtaposition raises questions about the feasibility of expecting CISA to effectively oversee AI safety initiatives when its own capacity to defend critical infrastructure is being reduced.
Anthropic: Praise, Partnership, and Pushback
The White House’s relationship with Anthropic exemplifies the policy’s contradictory nature. In April, Anthropic unveiled the Claude Mythos Preview, a model with advanced hacking capabilities that sparked concern over AI‑enabled threats. Shortly thereafter, Trump praised Anthropic’s leadership, telling reporters he “like[s] high‑IQ people, and they definitely have high IQs.” Yet the administration has also taken legal steps to bar Anthropic from most national‑security work, designating it a “supply chain risk” after a contract dispute over AI in warfare. When Anthropic attempted to expand access to Mythos for cyberdefense purposes—an activity aligned with the new executive order—the White House reportedly blocked the move, further muddying the administration’s stance toward the company.
China Policy: Deregulation, Chip Exports, and Data‑Center Tensions
Trump’s public justification for loosening AI regulation has repeatedly centered on maintaining an edge over China. He has championed slogans like “Build, baby, build” to encourage rapid data‑center construction and has argued that regulatory constraints hinder American competitiveness. At the same time, his administration permitted Nvidia to sell advanced AI chips to Chinese firms, rescinding an Biden‑era export control intended to impede Chinese AI progress. Simultaneously, after public outcry over rising electricity bills linked to data‑center expansion, the White House introduced a voluntary pledge for AI firms to adopt measures that would shield consumers from those costs—illustrating a tug‑of‑war between aggressive growth incentives and concessions to consumer concerns.
Public Opinion, Political Pressure, and the Limits of Symbolic Action
Observers suggest that the vacillations in Trump’s AI policy are driven partly by shifting public sentiment. As AI models have grown more capable, fears about job displacement, privacy, and concentration of power have intensified, prompting criticism from across the political spectrum—including figures as disparate as Steve Bannon and Bernie Sanders. The executive order allows the administration to appear responsive to these concerns without imposing substantive obligations on industry. By relying on voluntary cooperation, the White House can claim to be “doing something” about AI while avoiding the political risk of alienating Silicon Valley donors or appearing to impede the technological lead it frequently touts.
Conclusion: A Narrowing Window for Effective Regulation
With Anthropic, OpenAI, and their rivals amassing unprecedented economic and geopolitical influence, the opportunity for any single government to impose meaningful safeguards on AI is rapidly diminishing. The current executive order, while symbolically significant, lacks the teeth required to address pressing safety, security, and equity challenges posed by advancing AI systems. Unless future policy moves beyond voluntary agreements and embraces enforceable standards, the United States may find itself reacting to AI’s consequences rather than shaping its trajectory—a scenario that could undermine both national security and the broader public interest.