President Donald Trump dismissed calls to slow artificial-intelligence development during a trip to Ireland on September 13, arguing that the United States must retain its lead over China. He said AI guardrails could be put in place, but did not outline a proposal or announce a new policy.
The remarks arrive as a group of prominent AI executives and researchers has revived public arguments for slowing the advance of the most capable systems while safety measures catch up. They put the White House more plainly on the competitiveness side of that dispute: willing to discuss safeguards, but resistant to treating a broad pause in technical progress as the answer.
Competition first, with guardrails left undefined
Accounts from the BBC and Al Jazeera agree on the core of Trump’s message. He described the United States as ahead of China, said preserving that advantage was important and pushed back on the warnings behind the slowdown campaign. At the same time, he did not rule out guardrails.
That is a narrower position than rejecting oversight altogether. It is also far less specific than a regulatory program. The reports do not identify a new rule, executive order, timetable or type of safeguard attached to Trump’s comments. His intervention was political direction in a live policy argument, not a replacement for the administration’s existing AI orders.
The immediate dispute was prompted by calls associated with Anthropic chief executive Dario Amodei to reduce the pace of capability gains long enough to build stronger protections. Al Jazeera reported that OpenAI chief executive Sam Altman and xAI owner Elon Musk supported the warning. The BBC also reported competing views among people with AI-policy experience: former Anthropic and OpenAI researcher Jacob Coxon said an effective slowdown would need coordination with China, while White House AI and crypto adviser David Sacks said companies wanting a slower pace could choose one for themselves.
Those are positions in a policy fight, not a settled technical finding about AI risk or a demonstrated outcome of slowing development. The practical divide is over who should bear the cost of restraint: individual companies, Congress and federal agencies, or a coordinated international arrangement that would include strategic competitors.
What the White House has actually put in place
Trump’s comments fit more closely with two executive orders already issued by his administration than with a new federal licensing system. A December 2025 White House order on national AI policy called for a minimally burdensome framework and directed federal action against state AI laws judged inconsistent with that approach.
Then, on June 2, the administration issued its order on advanced AI innovation and security. Its design is more limited than a general requirement to obtain government permission before building or releasing an AI model. The order directs a voluntary framework for designated covered frontier models, with developers able to give the government access for up to 30 days before a release to other trusted partners.
The distinction is consequential. The June order calls for classified benchmarking to determine which model capabilities warrant coverage, and it assigns cybersecurity and national-security work involving agencies including the National Security Agency and the Cybersecurity and Infrastructure Security Agency. It also contemplates an AI cybersecurity clearinghouse. But its text explicitly says it does not authorize mandatory government licensing, preclearance or permitting for model development, publication, release or distribution.
In other words, the administration has paired security-oriented engagement around a selected class of frontier systems with a stated refusal to create a general federal gatekeeper for AI releases. Trump’s Ireland comments add a sharper China-competition rationale to that approach; they do not change the order’s voluntary structure.
A widening political split
Congress has not settled on a shared response. The BBC reported that House Speaker Mike Johnson warned against rushing AI regulation, while House Minority Leader Hakeem Jeffries argued for slowing development to protect the public. Those positions mirror the broader argument playing out among technology companies, researchers and officials: whether rapid deployment is chiefly a strategic advantage to protect, a safety problem to constrain, or both.
For companies developing frontier models, the difference between guardrails and preclearance is not semantic. A voluntary process can create channels for government testing and threat information without making a release conditional on approval. A licensing regime would give government a more direct role before models reach customers or partners. The June order expressly confines itself to the former category.
Trump has now made clear which side of the slowdown debate he favors, while retaining room for unspecified safeguards. The most concrete federal policy remains the June order’s combination of innovation, cybersecurity cooperation and voluntary access for a yet-to-be-designated set of frontier models—not a mandatory permit system for AI development.
