A reported dispute inside Donald Trump’s administration is putting two visions of AI policy against each other: David Sacks and technology executives favor a light-touch approach centered on innovation and corporate responsibility, while senior officials are pressing for stronger safeguards against cyberattacks and other national-security risks. The disagreement concerns oversight of advanced AI models, not a completed comprehensive federal AI law.
The latest account, published on September 16, 2026, describes senior officials meeting repeatedly over possible guardrails. A summary published on September 17 added details about concerns involving banking, water and power systems, as well as possible biological-weapons applications.
The White House split is about how far oversight should go
The light-touch position associated with Sacks holds that existing product-liability and fraud laws are largely sufficient for AI models. Sacks has also argued that companies themselves are responsible for maintaining control of their products. The broader argument is that additional government restrictions could slow innovation and weaken the United States’ position in the global AI competition.
The security-focused camp is concerned that advanced models could make cyberattacks more capable or easier to scale. Scott Bessent, the Treasury secretary, reportedly warned that AI-enabled attacks could seriously disrupt the banking system. White House chief of staff Susie Wiles was also reported to have pushed for stronger safeguards.
The dispute therefore reaches beyond familiar arguments about chatbot misinformation or workplace automation. It concerns whether the government should require more scrutiny before powerful models are deployed, particularly when failures could affect financial institutions or essential infrastructure.
| Policy priority | What it emphasizes | Risk it seeks to address |
| Light-touch oversight | Innovation, U.S. competitiveness and company responsibility under existing liability and fraud laws | Government rules could slow development and deployment |
| Security-focused oversight | Additional safeguards and longer government review of powerful AI models | Cyberattacks against banking and critical infrastructure, along with biological-security risks |
What happened to the planned review of advanced models?
In May 2026, David Sacks reportedly persuaded Trump to abandon a planned executive order that would have subjected powerful AI models to a longer government review. Wiles and Bessent were reportedly surprised and dissatisfied by that decision.
Trump later signed a narrower version after pressure from Wiles and Bessent, according to the reported account. The available description of that development does not include the order’s exact text or legal effect, so its specific requirements are not part of the policy picture here.
That chronology matters because it shows the dispute producing movement in both directions: a proposed review was dropped, then a more limited order was reportedly signed. The central fight is not settled by the existence of either step.
The security concerns driving the push for safeguards
Officials reportedly examined several ways advanced AI could create national-security problems:
- Banking: AI-enabled cyberattacks could disrupt financial systems.
- Critical infrastructure: Water and power systems were among the areas officials reportedly considered vulnerable.
- Cyber operations: More capable models could help identify software weaknesses or support sophisticated attacks.
- Biological security: Officials reportedly considered the possibility that AI could assist biological-weapons development.
These concerns explain why some officials have sought longer government review instead of relying entirely on companies to police their own systems. They also explain why the policy argument has moved from abstract questions about AI’s future to concrete worries about institutions that millions of people depend on.
How this fits Trump’s existing AI policy
The administration has already announced a national direction for AI policy, even as officials reportedly disagree over its safeguards.
On March 20, 2026, the White House released a national AI legislative framework with six objectives covering children and communities, intellectual property, free speech, innovation and workforce development. The framework argues for a uniform federal policy rather than a patchwork of conflicting state rules.
Earlier, on December 11, 2025, Trump signed Executive Order 14365. The order directs federal action against state AI laws considered inconsistent with a minimally burdensome national policy and calls for a federal legislative recommendation. It also preserves possible state authority in areas including child safety, data-center infrastructure and state government procurement.
That combination creates a two-level policy debate. The administration favors national uniformity in areas it believes should be handled federally, while the executive order leaves room for state action in selected fields. A single national framework does not automatically mean that every state AI rule would disappear.
The framework’s six objectives are:
- Protecting children and supporting parents.
- Addressing community concerns.
- Protecting intellectual property.
- Preserving free speech.
- Promoting innovation.
- Supporting workforce development.
The White House’s documents establish those policy aims. They do not establish a comprehensive federal AI regulation law replacing the ongoing debate over model oversight and safeguards.
Where technology executives fit into the dispute
The reported divide also has an industry dimension. Nvidia CEO Jensen Huang and Meta CEO Mark Zuckerberg were associated with opposition to additional regulation and coordinated slowdowns. Elon Musk was associated with support for stronger AI safeguards.
OpenAI and Anthropic executives were reportedly involved in discussions with officials about AI safety concerns. Anthropic’s Mythos model was described in the reporting as a warning sign because of its potential to help find weaknesses in government software and facilitate sophisticated cyberattacks.
Those positions do not form a single industry consensus. They reflect the same basic tension inside the administration: faster development and fewer new restrictions on one side, and more government scrutiny before deployment on the other.
The White House’s internal disagreement is therefore about more than whether AI should be regulated in the abstract. It is a fight over timing, responsibility and the threshold at which potential security risks justify government intervention.
