FTC Warning Puts AI Safety Exemption Under Scrutiny

The U.S. Federal Trade Commission’s chair has challenged a central premise of the latest AI safety debate: that frontier laboratories may need an antitrust exemption to coordinate on slowing advanced-model development.

Speaking at Georgetown University on September 15, FTC Chairman Andrew Ferguson said the public should be “deeply suspicious” when AI companies ask Washington for both extensive regulation and permission to cooperate outside normal competition rules. His warning was aimed broadly, but it follows Anthropic CEO Dario Amodei’s proposal for a narrow exemption allowing AI developers to coordinate on safety standards and the pace of development.

What changed

The intervention matters because it places the administration’s competition regulator directly against the industry’s emerging safety argument. AI companies have increasingly presented coordination as a public-interest necessity: if labs believe future systems could create catastrophic cybersecurity or biological risks, they argue, competitors should be able to share safeguards without fear of prosecution under antitrust law.

Ferguson offered a different interpretation. He said such requests could create “barriers to entry” that protect established companies from challengers. In that view, a safety exemption could become a mechanism for the largest laboratories to shape technical standards, limit competitive pressure and preserve their lead while presenting the arrangement as risk management.

Bloomberg Law reported that Ferguson characterized the proposal as “moat digging,” sharpening the concern that regulation could entrench rather than constrain the companies already controlling the most capital, computing capacity and distribution channels.

Why it matters

The dispute exposes a difficult policy collision. Some safety measures genuinely require cooperation: shared testing protocols, incident reporting, cyber-defense information and common thresholds for dangerous capabilities may be more effective if firms can coordinate. But a broad exemption could also let incumbents decide which competitors qualify, what development counts as “safe” and whether a slowdown applies evenly across the market.

That makes the FTC’s skepticism consequential for smaller model developers, cloud providers and investors. If Washington rejects a blanket waiver, companies may need to design narrower arrangements—such as government-supervised standards bodies, independently audited safety protocols or explicit safe harbors limited to technical information sharing. Those approaches could reduce competitive risk, but they would likely be slower and more bureaucratic.

The episode also complicates the political framing of AI regulation. Industry leaders have argued that government must act before systems become more capable, while critics suspect that some calls for regulation are partly efforts to raise rivals’ costs. Ferguson’s remarks give that critique official weight without resolving whether the underlying safety warnings are valid.

What remains uncertain

Ferguson stressed that he was expressing a personal view and that President Donald Trump will set the administration’s AI policy. The FTC has not announced a formal rule or enforcement action targeting the proposed coordination. It is also unclear whether Anthropic’s request would cover only safety research or extend to model release schedules, compute access and capability thresholds.

The next test will be whether policymakers can separate legitimate safety collaboration from market coordination. Without that distinction, the industry may face a choice between fragmented safety work and a regulatory bargain that makes the biggest AI companies even harder to dislodge.