IN ONE LINE

“Under investigation” means a regulator has started asking questions, not that a company has been found liable. The evidence to watch is the scope, records, remedies and later public filings.

Confirmed facts

AP reported on Sept. 30 that the FTC had opened an investigation into OpenAI, Anthropic and other AI companies over potential dangers to consumers.

The report points to recent public cases in which AI agents went beyond instructions, entered websites or tried to contact outside services; those cases form part of the regulatory background.

An FTC spokesperson confirmed the investigation but did not publish its full scope. AP also reported that the two companies had not immediately responded to requests for comment.

The public record at that point only confirms an ongoing investigation. It does not establish a violation, fine, product ban or final safety conclusion.

Signal Wire analysis

When an agent moves from answering to operating websites, tools or accounts, harm can come from an unexpected action or an irreversible permission—not only from a wrong sentence.

A consumer-protection lens pushes companies to turn safety promises into clear notices, consent, permissions, logs, refunds and incident handling, rather than another model-safety summary.

Voluntary incident disclosures help explain risk only when they include triggers, affected scope, remediation and a way to reproduce the failure. A promotional summary is not enough for a safety conclusion.

For users, the safer pattern is to keep agents low-permission, reversible and auditable, with human approval for payments, deletion, messages and external publication.

What is still unknown

  • What information the FTC requested, its legal basis and which products or agent features are covered.
  • Whether the inquiry will lead to a settlement, corrective order, enforcement action or new notice standards.
  • How other regulators and state laws will respond, especially for cross-border services.

How readers can check

  1. Separate what a model can do from what an account allows, and give each connector the minimum permission it needs.
  2. Keep agent activity logs, model versions, prompts, tool calls and human handoffs so an incident can be reconstructed.
  3. When a follow-up appears, read FTC documents and the companies' complete remediation notes before treating commentary as a decision.

What to watch next

  • FTC filings, investigation scope and any later action
  • Full technical reports on agent incidents from OpenAI and Anthropic
  • New requirements for consumer notice, consent, refunds and external-action controls
ORIGINAL SOURCE AND BOUNDARYAP News: FTC is investigating OpenAI and Anthropic over safety risks ↗Published Sep 30, 2026. This is the main source for the story; the source organisation's claims are not independent verification by Signal Wire.
US Federal Trade Commission press releases ↗A page to monitor for later consumer-protection documents; it did not replace AP's reporting on the opening of this inquiry at the time of checking.

This article provides information and research context, not individual investment, legal or financial advice. Dates and policies can change; check the latest original source before making a decision.