The FTC asked seven companies how their companion bots are tested

The conversation

On 11 September 2025 the Federal Trade Commission announced it had issued orders to seven companies operating AI chatbots that can act as companions: Alphabet, Character Technologies, Instagram, Meta Platforms, OpenAI OpCo, Snap, and X.AI. The Commission used its Section 6(b) authority, which its own release describes as the power to "conduct wide-ranging studies that do not have a specific law enforcement purpose," to ask how these companies test, monitor, and try to limit harm to children and teenagers who use the products.

What the documents show

The FTC's release states the orders seek information across eight areas: how companies monetize engagement with the chatbots; how they process user input and generate outputs; how chatbot characters are developed and approved; what safety testing happens before and after launch; what steps reduce negative effects, particularly for minors; what is disclosed to users about features, audience, and data practices; how community guidelines and age restrictions are enforced; and how personal information gathered in conversations is collected, used, and shared. A separate FTC record, the report listing for the 6(b) orders, confirms the study's formal title concerns "Advertising, Safety, and Data Handling Practices" and dates it to September 2025. Commissioners Melissa Holyoak and Mark Meador each issued separate public statements alongside the order, which the press release notes without quoting.

The system boundary

The Commission's own framing draws the boundary between this action and an enforcement case: a 6(b) study compels information without alleging any company broke the law. The order does not find that any of the seven companies failed to protect minors; it asks each to describe, in its own records, what safeguards and monitoring already exist. Whatever a company's product does when a conversation touches self-harm, romance, or a child user, the inquiry's own documents describe only the request for information, not a judgment on the answer.

Where it fails

Because a 6(b) study has no law-enforcement purpose, its own materials give no timeline for when, or whether, the FTC will publish findings or take further action, and the information request does not itself constrain what a company's chatbot can say to a minor while the study proceeds. An operator or builder in this space should treat the order's eight categories as a preview of the questions a future regulator, plaintiff, or state attorney general might ask.

• Could we answer, with documentation, each of the eight categories the FTC named, for our own product?
• Do our safety-testing records distinguish pre-launch testing from ongoing post-launch monitoring, as the order's own language separates them?
• What does our age-restriction enforcement log actually show, rather than what our public policy states?

This is an editorial reading of a study's request, not of any conclusion; the FTC's own materials record no findings as of this writing.