In May 2026, a California resident named Amargo Couture sued OpenAI. Her allegation is narrow and, if true, ordinary: she used ChatGPT, she typed in things about her health and her finances, and information about that activity was transmitted to Meta and Google through embedded third-party tools — the Meta Pixel and Google’s analytics code.

In August, OpenAI moved to dismiss in the Southern District of California, before Judge Marilyn Huff. The core of the motion is one sentence:

All of plaintiff’s claims are barred because she consented to OpenAI’s terms of use and privacy policy, which expressly inform users that information about their use of ChatGPT, including their prompts, may be disclosed to third parties.

That is it. Not “this did not happen.” Not “the data was anonymised.” You agreed.

Why the Argument Is Legally Serious

It is easy to dismiss this as lawyering. It should not be. Consent is a complete defence to most of the claims typically stacked in these cases — the wiretap theories in particular turn on whether a party to the communication authorised the interception, and a privacy policy that discloses third-party sharing is the standard way defendants establish that.

The pixel-litigation wave of the last three years has repeatedly run into this wall. Where the policy is specific, courts have generally found consent. Where the policy is vague or contradicted by marketing claims, plaintiffs have survived dismissal. That is why newer complaints, as practitioners have noted, increasingly bolt consumer protection claims about misrepresentation onto the interception claims — it is easier to argue a company lied about its privacy practices than that a disclosed practice was unauthorised.

So the motion may well work. The question worth asking is what it means that it might.

Consider what OpenAI is describing as an agreement.

A person opens a chat interface. There is no visible mention of Meta or Google. There is a link to a privacy policy, which almost nobody reads, and which — like nearly every privacy policy — says that information “may be” shared with “third parties” including “analytics providers.” The person types a question about a symptom, or a debt.

Nothing in that sequence resembles what the word consent means in any other context. There was no specific disclosure of the recipients. There was no choice presented at the moment the data was generated. There was no option to use the product without the analytics tags. And the thing being consented to — the content of what you type into an AI assistant — is categorically unlike the browsing metadata these clauses were written for.

That last point is the one that matters most, and it is why this case is different from the hundred pixel cases before it.

A Prompt Is Not a Pageview

Website analytics were built to capture behavioural exhaust: which page, how long, where from. Privacy law has treated that as low-sensitivity for thirty years, and the boilerplate consent language grew up around that assumption.

A prompt is not exhaust. A prompt is a disclosure.

People tell chatbots things they do not tell their doctor, their partner, or their search bar — because the interface behaves like a conversation and the model answers like a confidant. We wrote about the same dynamic when therapy transcripts started appearing in court and again when OpenAI began age-predicting teenagers.

Applying pageview-era consent doctrine to prompt-era disclosure is a category error. It is legally coherent and substantively absurd, which is a fairly precise description of most of American privacy law.

Why Third-Party Tags Are There At All

Worth stating plainly: the Meta Pixel and Google analytics tags on a consumer web product are marketing infrastructure. They exist to measure conversion, attribute signups to ad spend, and build retargeting audiences. That is their function; it is not an accident or a misconfiguration.

Which means the disclosure at issue, if it occurred, was not a leak. It was the product working as instrumented, and the dispute is only about whether the paperwork covered it.

What This Case Will Actually Decide

Not very much, unfortunately — and that is the point. If Judge Huff grants the motion, the holding will be that a privacy policy disclosing third-party sharing defeats consent-based claims. If she denies it, the case proceeds to discovery about what was actually transmitted, which would itself be valuable: nobody outside these companies knows what fields the tags on an AI chat interface actually capture.

Either way, the underlying arrangement is untouched. Consent-as-defence does not require anyone to stop; it requires them to have written it down.

The regimes that would produce a different answer are the ones that treat consent as something you have to give rather than something you fail to refuse: GDPR Article 7’s requirement that consent be freely given, specific, informed and unambiguous, and the specific-consent requirements now live in India’s DPDP rules. Under those frameworks, “the policy said prompts may be disclosed to third parties” is not consent. It is notice, and notice is not the same thing.

What To Do

  1. Assume anything you type into a hosted chatbot may be instrumented. Not stolen — instrumented, by design, and disclosed in a document you did not read.
  2. Turn off chat history and model training where offered. In ChatGPT this is under Data Controls. It does not affect third-party analytics tags, but it reduces retention.
  3. Block the tags. uBlock Origin or Brave shields stop the Meta Pixel and analytics beacons from firing regardless of what any policy permits. This is the only step here that does not depend on a company’s promise.
  4. Use a browser container or a separate profile for AI tools. It breaks the join between your chatbot session and your logged-in Meta and Google identities, which is the join that makes the data valuable.
  5. Do not put identifiable health or financial specifics into a hosted assistant. For genuinely sensitive questions, a locally-run open-weights model is now good enough for most tasks — we covered the current landscape — and it sends nothing anywhere.
  6. Read the sentence, not the policy. If you read nothing else, search any privacy policy for the words “third parties” and “analytics.” That is where the consent you are giving actually lives.