AI Disclosure vs AI Consent: The Two Duties Websites Keep Confusing
There's a quiet mistake spreading across the web in 2026: businesses add "this is an AI assistant" to their chatbot and believe they've handled AI compliance. They've handled disclosure — one of two separate duties. The other is consent: whether that AI may process a visitor's data before they've agreed. And that second duty is turning into the next website-litigation wave, because AI chat is the new session replay. Here's the map, and why one notice isn't enough.
Key takeaways
- Disclosure and consent are different duties. Telling visitors "you're talking to an AI" satisfies a disclosure law; it says nothing about whether the AI may process their data before they agreed.
- AI on a website is actually three layers: labeling AI-made content, disclosing an AI interaction, and consenting to AI data processing. Most sites handle only the first two — and the third is where the risk is moving.
- AI chat is the new session replay. The CIPA wiretapping theory that hit session-replay tools has migrated to AI chatbots, and it's now the fastest-growing category of privacy litigation.
- SB 690 doesn't stop it. California's 2026 reform narrows the pixel/pen-register theory but leaves the §631 wiretapping provision intact — so the AI-chat suits continue.
- The fix is order of operations. Disclose the AI, get consent before the conversation flows to a vendor, and keep proof — the same prevention-first move that answered the tracking wave.
Disclosure vs consent: the distinction everyone blurs
Start with the mistake, because naming it clearly is most of the fix. When a business puts "I'm an AI assistant" in its chatbot's first message, it has done something real and required — but it has answered only one question. There are two.
AI disclosure answers: does the visitor know they're dealing with AI? It's about honesty — nobody should be fooled into thinking a machine is a person. This is what the EU AI Act's Article 50 and a wave of US chatbot laws require, and it's satisfied by a clear notice.
AI consent answers a different question entirely: may this AI process the visitor's data before they've agreed? It's about the data flow — when your chatbot ships what a visitor types to a third-party AI vendor, that's processing personal data, and both the GDPR and California's wiretapping law care whether the visitor consented first. A perfectly worded "I'm an AI" notice does nothing to answer this. You can disclose flawlessly and still have a consent problem.
That gap — disclosure handled, consent ignored — is the exposure quietly opening up across the web. And it maps onto a bigger picture worth having in full.
The three layers of AI on a website
"AI compliance" is really three duties wearing one name. Keeping them straight tells you which apply to you, and which tool handles which.
Layer A — content labeling. Watermarking AI-generated images, video and audio so people aren't fooled by synthetic media. This is California's SB 942 (the AI Transparency Act) and Article 50(2) of the EU AI Act. Important reality check: SB 942 only applies to large generative-AI providers above a one-million-California-user threshold, and excludes text — so it rarely touches an ordinary website. If you're a Shopify store or an SMB, this layer usually isn't yours.
Layer B — interaction disclosure. Telling visitors they're interacting with AI. This applies to any site with a chatbot, and it's the disclosure duty above. We go deep on it in EU AI Act Article 50 for websites.
Layer C — data-processing consent. Getting consent before the AI processes a visitor's data. This is the consent duty — and it's the one turning into litigation. Which is where the story gets specific.
Why AI chat is the new session replay
If you watched the website-tracking lawsuit wave, you already know this pattern — because it's the same one, aimed at a new target. For three years, plaintiffs' firms sued websites under California's Invasion of Privacy Act (CIPA) over session-replay tools and pixels that intercepted a visitor's activity before consent. The theory: when a third-party vendor gets real-time access to what happens on your site, that vendor is an uninvited eavesdropper on a private communication — an unlawful "interception" under CIPA §631.
In 2025 and 2026, that exact theory migrated to AI chatbots. When a visitor types into your chat widget and the widget ships that message to a third-party AI vendor in real time, plaintiffs argue the structure is identical — the vendor intercepts the communication "in transit," the website operator "aided" it, and no one got the visitor's consent first. Courts have found this viable at the pleading stage, and the numbers tell the story: chatbot wiretap filings grew from a couple of matters in 2021 to dozens by the end of 2025, and legal commentators now call it the fastest-growing category of privacy litigation. The damages are the same as the session-replay wave — $5,000 per violation under Cal. Penal Code §637.2, a private right of action, pleaded per session.
What decides these cases is worth memorizing, because it's also the fix. Courts and defense counsel keep returning to two questions: (1) Did the visitor consent before the conversation began? And (2) Can the vendor use the conversation for its own purposes? A site that can answer "yes, consent came first" and "no, the vendor is contractually limited" is in a materially stronger position than one relying on a disclosure notice. Notice the through-line to the tracking wave: it was never really about the technology — it was about whether the interception happened before the visitor agreed.
Is your AI chat firing before consent?
The consent question is a technical one: does your chat widget transmit the conversation to a vendor before the visitor agrees? Our free scanner now detects AI surfaces alongside trackers — see what's on your site in about 10 seconds, no account.
Scan your site free →The disclosure landscape (Layer B)
Even setting litigation aside, the disclosure duty alone has multiplied into a patchwork worth knowing. The rule everywhere is the same in spirit — tell people it's AI — but the specifics and the teeth vary.
| Law | What it requires | Teeth |
|---|---|---|
| EU AI Act Art 50(1) | Disclose AI interaction at the first interaction (live 2 Aug 2026) | Up to €15M or 3% turnover |
| California SB 1001 | Bots must disclose they're bots (since 2019) | State enforcement |
| California SB 243 | Companion chatbots disclose non-human status + minor safeguards | Private suit: $1,000/violation + fees |
| New York (GBL Art 47) | Companion AI disclosure + crisis protocols | AG: up to $15,000/day |
| Washington HB 2225 / SB 1546 | Chatbot non-human disclosure (eff Jan 1 2027) | Private suit (statutory damages) |
| Colorado AI Act | Disclose AI interaction unless obvious (eff Feb 2026) | State enforcement |
Two things stand out. First, several of these carry a private right of action — California's SB 243 and the Washington and Oregon chatbot laws let individuals sue directly, which is what turned AI disclosure from a policy footnote into real exposure. Second, they stack: an EU-facing site with a chatbot can owe an Article 50 disclosure, a US state disclosure, and — separately — a consent obligation. Handling one doesn't handle the others. But disclosure, as thorough as this list is, is still only half the problem.
The consent problem (Layer C)
Here's what makes the consent layer harder than disclosure, and why a notice can't stand in for it. Disclosure is a message — you show it and you're done. Consent is a gate — it has to actually hold the AI back until the visitor agrees, and then let it through. A disclosure that appears while the chat widget has already loaded and transmitted the first message hasn't gated anything; it's a sign on an open door.
Meeting the consent duty properly means a specific sequence:
This is precisely the shape of what we built into ConsentPixel's AI consent management: a dedicated ai_processing consent purpose that holds AI chat assistants until the visitor opts in, a one-step disclose-and-consent modal, and immutable notice logging anchored externally for proof. We're candid about the boundary — it governs AI tools that process visitor data (chat assistants, LLM endpoints), not the watermarking of AI-generated content, which is Layer A and a different job. This is information, not legal advice, but the durable position is the same one that answered the tracking wave: consent before the data flows, provable after the fact.
What to actually do
The three layers collapse into a short, ordered checklist. Most websites can skip Layer A entirely.
Frequently asked questions
What's the difference between AI disclosure and AI consent?
AI disclosure means telling visitors they're interacting with AI — a message, required by the EU AI Act's Article 50 and a wave of US chatbot laws. AI consent means getting the visitor's agreement before that AI processes their data — a gate, required in effect by the GDPR and California's wiretapping law when a chatbot ships a conversation to a third-party vendor. Disclosure is satisfied by showing a notice; consent requires actually holding the AI back until the visitor opts in. You can disclose perfectly and still have a consent problem, which is exactly the gap most sites miss. This is general information, not legal advice.
Why is AI chat called "the new session replay"?
Because it's the same lawsuit theory pointed at a new target. Plaintiffs spent years suing websites under California's CIPA over session-replay tools that intercepted visitor activity before consent. In 2025–2026 that theory migrated to AI chatbots: when a chat widget transmits a visitor's messages to a third-party AI vendor in real time, plaintiffs argue it's an unlawful interception under Section 631, with statutory damages of $5,000 per violation under Cal. Penal Code §637.2. Chatbot wiretap claims grew from a couple of filings in 2021 to dozens by the end of 2025 and are now described as the fastest-growing privacy-litigation category. This is general information, not legal advice.
Does disclosing that my chatbot is AI protect me from a lawsuit?
It helps with the disclosure duty, but it doesn't address the consent duty — and the lawsuits are about consent. Telling a visitor "I'm an AI" satisfies laws like SB 243 and the EU AI Act's Article 50, but it says nothing about whether the chatbot may transmit the conversation to a third-party vendor before the visitor agreed. Under CIPA, that pre-consent interception is the violation, regardless of how well the AI is labeled. The protective move is to hold the AI until consent, limit what the vendor can do with the conversation, and keep proof — not just to add a notice. This is general information, not legal advice.
Does California's SB 942 (AI Transparency Act) apply to my website?
Probably not, unless you're a large generative-AI provider. SB 942 applies to businesses that make a publicly accessible generative-AI system producing images, video or audio, with at least one million monthly California users — and it excludes text-only AI. That scopes it to big AI providers, not ordinary websites or Shopify stores. It requires content labeling, embedded provenance data, and a free public detection tool, enforced by the Attorney General with no private right of action. Most websites should focus instead on disclosing their chatbot (interaction) and gating what it does with visitor data (consent), which are the duties that actually apply to them.
How do I get consent for an AI chatbot without losing the chatbot?
Hold it, don't hide it. The chatbot's launcher can still appear, so visitors know help is available — but the widget doesn't load or transmit until the visitor engages it and agrees. A one-step prompt that both discloses it's AI and asks to enable it resolves disclosure and consent together, before any data flows; if the visitor declines, the AI simply stays held. For a business's own AI code, a consent gate lets the code wait for the opt-in. You keep the AI you rely on; it just waits for permission. This is general information, not legal advice.
Is AI consent required under the GDPR too, or just CIPA?
Both regimes point the same way. Under the GDPR, using AI to process an EU visitor's personal data needs a lawful basis, and for a chatbot that basis is usually consent — freely given, specific, informed, and obtained before processing. Under California's CIPA, the concern is interception before consent. They come from different legal traditions but converge on the same behavior: don't let the AI process a visitor's data until they've agreed. Solve that one thing and you address both. Whether either applies depends on your visitors and your AI use; this is general information, not legal advice.
The bottom line
AI on a website is two duties, not one — and the confusion between them is the exposure. Disclosure tells people it's AI; consent controls whether the AI may process their data before they've agreed. A notice handles the first. Only a gate handles the second.
And the second is where 2026 is heading. The CIPA wiretapping theory that made session replay a liability has moved to AI chat, it's the fastest-growing privacy-litigation category, and California's reform leaves that exact theory intact. The businesses that stay out of trouble will be the ones that treated AI chat like session recording before the demand letters arrived: disclose it, gate it, prove it.
That's not a new playbook. It's the prevention-first move that answered the last wave — pointed at the next one.
Disclose and gate the AI on your site
ConsentPixel — Privacy · Verified surfaces the Article 50 disclosure, holds AI chat until the visitor opts in, and logs every decision as verifiable proof — the CIPA-grade consent layer for AI. Start with a free scan, then a 14-day trial.
Scan your site free →Information, not legal advice. This article explains AI disclosure and AI consent obligations for general educational purposes and does not constitute legal advice or create an attorney–client relationship. How the EU AI Act, CIPA, US state chatbot laws, SB 942, and the GDPR apply to your site depends on your specific AI use and visitors, and these laws — and the case law described here — are evolving and, in places, unsettled. Dates and penalties reflect publicly reported information as of September 2026 and may change, including the status of California's SB 690. The $5,000-per-violation figure reflects statutory damages under California Penal Code §637.2. Consult qualified counsel for your situation. ConsentPixel — Privacy · Verified is not a law firm and does not make any website compliant by itself; it does not watermark AI-generated content.