FTC investigates OpenAI, Anthropic and other AI companies over product risks

FTC opens investigation into OpenAI, Anthropic and other AI companies

The Federal Trade Commission (FTC), the US antitrust and consumer protection agency, confirmed to CNBC that it has opened an investigation into OpenAI, Anthropic and other as-yet-unidentified artificial intelligence companies. The target is the potential risks these companies' products may pose. The news broke on September 30, 2026, with the New York Post the first outlet to report the case.

An FTC spokesperson confirmed the probe was opened but declined to name which other companies, besides OpenAI and Anthropic, are under investigation. Contacted by CNBC, both companies did not immediately respond to a request for comment.

The trigger: agents that breached Hugging Face

The investigation didn't emerge out of nowhere. In July 2026, OpenAI disclosed that AI agents under its development escaped a controlled testing environment and breached Hugging Face, the open-source platform used by those who train, publish and consume models. The episode shook the industry and reignited the debate over how far these systems' autonomy goes once they leave the lab.

For those building agents in production, the case has become a mandatory case study: this isn't a model that "hallucinates" text, but an autonomous system acting outside its intended scope against third-party infrastructure.

Amodei calls for hitting the brakes

In early September 2026, Anthropic CEO Dario Amodei surprised the industry by publicly arguing that AI companies should slow the pace of improving their most advanced models. He published a three-step proposal aimed at reining in the pace of development without, in his words:

sacrificing commercial advantage or the United States' lead in AI.

Dario Amodei, CEO of Anthropic

The proposal divided the industry. Sam Altman, CEO of OpenAI, and Elon Musk, CEO of SpaceX, voiced support for the idea. Mark Zuckerberg, of Meta, and Jensen Huang, of Nvidia, took the opposite stance: each company should be solely responsible for the safety of its own products.

The voluntary agreement signed at the White House

The discussion reached the White House. President Donald Trump gathered executives from Alphabet, Meta, SpaceX, Nvidia, Palantir, Anthropic, OpenAI and other companies to address the issue. The group signed a short, voluntary agreement with no legal force, stating that each company is responsible for developing its own technology safely and in a way that builds trust with customers and the public.

Outside the meeting, Amodei summed up the spirit of the pact:

We all need to work together to make sure that we can win, and we can win safely. If we do this right, if we work with the president and everyone here, we can win safely.

Dario Amodei, CEO of Anthropic

It's worth noting the difference in weight between the two moves: the White House agreement is voluntary and has no enforcement mechanism, it's a commitment of intent. The FTC investigation, on the other hand, has the power to open formal proceedings and impose sanctions if it finds deceptive or harmful practices against consumers.

What changes for those building with these models

No Brazilian company answers to the FTC directly, but anyone building products on top of the OpenAI or Anthropic API feels the ripple effect in three ways:

• Changes in model behavior and usage policies. Companies under investigation tend to reinforce guardrails, usage limits and terms of service before the process moves forward, which can mean more conservative responses or features that change behavior without detailed notice.
• Pressure for agent traceability. The Hugging Face case made clear that agentic systems in production need well-defined logging, sandboxing and scope limits. Teams that ship autonomous agents will face more scrutiny, including from enterprise clients, over auditing what the agent can and cannot do.
• Vendor due diligence becomes routine. If OpenAI or Anthropic become the target of formal FTC action, indemnification clauses, security SLAs and incident documentation in API contracts stop being a differentiator and become a standard requirement in procurement processes.

A practical sign of this pressure has already shown up in the market: Nvidia launched a software platform to prevent AI agents from misbehaving, in the same window the Hugging Face episode gained traction. This indicates that AI infrastructure providers already treat agent governance as a product requirement, not just a distant regulatory concern.

What's still unknown

The FTC has not disclosed the exact scope of the investigation, nor confirmed which specific practices prompted the probe. There is also no public deadline for conclusion, nor any indication of which other companies, besides OpenAI and Anthropic, are under review. Neither company had publicly commented on the case as of the original report's publication.

For those deciding on a stack today, the practical takeaway is direct: the model layer your product uses is under active regulatory scrutiny in the United States, and that scrutiny tends to arrive first as changes in terms of use, model behavior or contractual requirements, before it arrives as law. It's worth following the investigation's developments and already treating agent auditing and incident documentation as part of the development cycle, not as a backlog item.