US President Donald Trump has signed a new executive order creating a voluntary mechanism under which developers of the most advanced artificial intelligence (AI) models can give the US government early access to their systems before public release. The move marks a notable shift for an administration that has generally favoured a hands-off approach to the powerful technology.
Under the executive order, titled ‘Promoting Advanced Artificial Intelligence Innovation and Security’, federal agencies are directed to establish a framework through which AI developers can determine whether their systems qualify as “covered frontier models” and, if they choose to participate, provide government agencies access to those models for security testing before wider deployment. The White House said the objective is to strengthen cybersecurity and help protect critical infrastructure from vulnerabilities that increasingly capable AI systems may uncover or exploit.
An earlier draft of the executive order had proposed giving federal agencies up to 90 days to evaluate frontier AI models before their public release, but the final version signed by Trump reduced the review window to 30 days following concerns from industry and administration officials that a three-month process could slow innovation and weaken the competitiveness of US AI companies.
The executive order comes as powerful AI models like Anthropic’s Mythos become capable of exploiting cyber vulnerabilities at an unprecedented scale, and as the Pentagon pushes to integrate frontier AI into military operations.
Former White House AI and cryptocurrency czar David Sacks said that while he understood concerns stemming from the order on account of bureaucratic mission creep, the change in review timelines was a “game changer”.
“The change in the EO (executive order) from a 90 day to 30 day period is a game changer because it allows our AI labs to comply with the voluntary framework without delaying new model releases. They can synchronise their efforts under the EO with other pre-release activities,” Sacks said on social media platform X.
Anthropic said that the executive order was “an important step in strengthening America’s leadership in AI,” and that it was looking forward to “collaborating with the White House to support its implementation”.
Story continues below this ad
In 2024, India had attempted a similar move, only to go back on it just weeks later due to significant pushback from the tech industry. Just ahead of the Lok Sabha elections that year, the IT Ministry issued an advisory requiring AI platforms to seek government permission before launching “under-testing, unreliable” services, but revoked it a fortnight later, owing to mounting criticism.
What does Trump’s AI executive order say?
One of the key provisions is a 30-day review period. Participating developers would allow designated government agencies to test advanced models for up to a month before they are released outside government circles. Agencies including the Departments of Defense, Commerce, Homeland Security and Treasury are expected to be involved in the process. The order also calls for stronger cyber defences across government systems and greater coordination with industry on identifying software vulnerabilities.
The framework remains voluntary. The executive order explicitly avoids creating any new licensing regime or mandatory approval process for AI models. Agencies are to design a “voluntary framework with AI developers” through which companies can engage with the federal government regarding frontier models.
The final version is significantly narrower than an earlier draft that the White House had prepared in May. That proposal would have given the government up to 90 days to review frontier AI systems before release. Trump abruptly postponed signing that version after concerns emerged that such a long review period could handicap American AI firms in their competition with China. According to reports, administration officials and industry advisers argued that a three-month review window was too burdensome for a sector where model releases occur at a rapid pace.
Story continues below this ad
The revised order cuts the review period to 30 days while retaining the core cybersecurity objectives. Trump had publicly indicated that he did not want any policy that could undermine America’s lead in AI, particularly against Chinese rivals. The shorter timeline is therefore seen as an attempt to balance national security concerns with the administration’s broader goal of maintaining US technological leadership.
“It (the order) reflects an administration trying to sustain its deregulatory, innovation-first posture while confronting the novel cyber risks posed by powerful new tools like Anthropic’s Claude Mythos Preview,” wrote Matthew Ferren, an international affairs fellow in national security at the Council on Foreign Relations.
“The order may yield short-term cybersecurity benefits, but its long-term effect on the marginal costs of compromise is less clear. Ransomware gangs may still find it cheaper to buy stolen credentials or off-the-shelf malware than to develop bespoke AI-driven exploit chains, and the most capable state actors, including China, already find it easy to penetrate poorly defended critical infrastructure networks,” he added.
When India attempted a similar move
On March 1, 2024, just months ahead of Lok Sabha elections, the IT Ministry issued an advisory requiring AI platforms to seek government permission before launching “under-testing, unreliable” services, only to go back on it a fortnight later. Faced with a strong backlash, the government had first clarified that its advisory was directed towards “significant” platforms and not startups before it revoked the requirement entirely.
Story continues below this ad
Both publicly and privately, it was facing a barrage of criticism. Within a week, on March 7, 2024, the National Association of Software and Service Companies or Nasscom, India’s largest tech industry lobby group, which also represents tech companies like Meta, Google, Amazon, Microsoft, and other software as a service (SaaS) companies — wrote to the IT Ministry, and made three key demands: relax applicability norms to limit the scope of the advisory, remove the contentious requirement for AI companies to seek government permission, and drop the need for companies to prepare an action taken and status report on account of compliance, according to correspondence reviewed by The Indian Express as part of a Right to Information (RTI) application.
Aravind Srinivas, founder of Perplexity AI, had called the advisory a “bad move by India”, while Martin Casado, general partner at the US-based investment firm Andreessen Horowitz, had termed the move a “travesty”, which was “anti-innovation” and “anti-public”.
