# What exactly is the FTC investigating?

The New York Times reported that the agency will examine whether the AI labs have broken federal laws prohibiting companies from unfair and deceptive practices. That phrase maps to Section 5 of the FTC Act, the same statute the agency has used against everything from false weight-loss claims to deceptive data-security promises. In an AI context, the theory is straightforward: if a company tells consumers a model is safe, reliable, or aligned, and the evidence does not support that, the gap is a deceptive practice. The source material does not name specific claims under review, and that omission is itself informative. The FTC typically opens a probe with a broad civil investigative demand and narrows later. According to the New York Times, the investigation is at the examination stage — the agency is determining whether laws were broken, not alleging that they were. My read: the target is the marketing layer, not the model weights. Regulators have struggled to regulate AI capability directly. They have decades of muscle memory regulating advertising claims. This probe is the FTC doing what it already knows how to do.

# Why OpenAI and Anthropic specifically?

Both companies occupy the same rhetorical space: frontier capability paired with explicit safety commitments. OpenAI publishes system cards and a preparedness framework. Anthropic publishes a responsible-scaling policy and has made safety part of its public identity. That combination — high-stakes claims plus high visibility — is exactly what draws consumer-protection attention. According to the New York Times, the agency's focus is potential consumer harms, which suggests the investigation is centered on what users were told and what they experienced, not on abstract existential risk. The practical implication is that internal safety evaluations, red-team reports, and marketing copy all become discoverable.

# How do the two companies compare under this scrutiny?

Dimension	OpenAI	Anthropic
Primary public safety claim	Preparedness framework and system cards	Responsible-scaling policy and constitutional AI
Consumer surface area	ChatGPT, largest consumer AI user base	Claude, smaller consumer base, heavier enterprise mix
Likely document exposure	Very high — years of public capability claims	High — safety positioning is central to brand
Regulatory posture to date	Engaged, sometimes adversarial with regulators	Engaged, positions itself as pro-regulation
Verdict	OpenAI has more surface area to defend; Anthropic has more brand equity riding on the outcome. Both lose if the probe forces disclosure of internal safety disagreements.

# What does the FTC actually get out of this?

The agency's leverage is not fines alone. Section 5 enforcement can produce consent orders that dictate how a company describes its products for years, with compliance reporting requirements attached. That is a more durable outcome than a one-time penalty. The New York Times reported the investigation is focused on potential consumer harms, which keeps the theory of harm grounded in user experience rather than speculative future risk. That framing matters because it is the version of AI regulation most likely to survive a court challenge: it borrows from established advertising and consumer-protection doctrine rather than inventing new AI-specific rules.

# What should the rest of the industry expect?

Every consumer-facing AI company should assume its marketing language is now in scope. The FTC's action against two leaders signals a template. Startups that describe models as 'safe,' 'hallucination-free,' or 'enterprise-grade' without documentation are exposed to the same theory. The second-order effect is a disclosure chill. Companies will start writing marketing copy the way public companies write risk factors — hedged, qualified, and reviewed by counsel. That is good for consumers and bad for anyone whose differentiation was a confident claim.

# Predictions

1. By Q2 2027, the FTC will issue a civil investigative demand to at least one additional foundation-model company beyond OpenAI and Anthropic, expanding the probe to a third major lab. 2. By mid-2027, OpenAI or Anthropic will enter a consent order restricting specific marketing language around safety and capability claims, with compliance reporting for at least three years. 3. By end of 2027, at least two state attorneys general will open parallel unfair-and-deceptive-practices investigations into AI labs, using the FTC action as a template.

# Article summary

• The FTC investigation is a consumer-protection action, not a novel AI regulation — which makes it far more legally durable than earlier AI-specific proposals.
• The real exposure for OpenAI and Anthropic is discovery: internal safety debates and unpublished evaluations become evidence.
• Anthropic has more brand equity at risk; OpenAI has more consumer surface area to defend.
• Every consumer-facing AI company should now treat safety and capability marketing as a regulated claim.
• The likely end state is a consent order that constrains language, not a fine that constrains behavior.