The EU AI Act's December 2026 AI-Content Deadline: An Outbound Playbook

What actually changes on December 2

I'm Hlib Storchak. I build outbound systems for B2B founders and sales teams, and I've booked 2000+ meetings for B2B clients doing it. Compliance dates like this one land on my desk more often than they used to, because a growing share of the outbound stack I touch now involves some form of generative AI: an AI avatar recording a LinkedIn video, a voice-cloning tool reading out a script, or a copywriting assistant drafting the first line of a sequence.

The short version: nothing new is being introduced on December 2, 2026. The EU AI Act's Article 50 transparency rules, including the duty to mark AI-generated content so it can be detected as artificially generated, have applied since August 2, 2026. What lands on December 2 is the end of a narrow grace period that applies only to the machine-readable marking requirement, and only for AI systems that were already on the market before August 2, per the European Commission's own FAQ on Article 50 transparency obligations.

If you are not a vendor building generative AI tools yourself, this is mostly a question of who else's compliance you are relying on. That is the part most outbound teams have not actually checked yet.

The two dates that matter, and why they are different

It is easy to conflate these, and I have seen client teams do it. They are not the same obligation.

August 2, 2026: Article 50 as a whole applies from this date. That covers telling people they are talking to an AI, disclosing deepfakes, and labeling AI-generated text published to inform the public on matters of public interest, per the practical guide to Article 50 from the EU AI Act's public reference site.

December 2, 2026: A four-month grace period ends for one specific piece: the machine-readable marking of AI-generated audio, image, video and text output, under Article 50(2), and only for generative AI systems that were already placed on the market before August 2, 2026. Any system entering the market on or after August 2 gets no such grace period, it has to mark from day one.

Who carries the obligation: providers vs deployers

This is the part that matters most for a B2B team, because most of you are not the ones the marking duty is aimed at.

The Article 50(2) machine-readable marking duty sits with providers, meaning the company that builds the generative AI system: the AI video tool, the voice-cloning platform, the avatar SDR product. If you are a business using one of those tools to run outbound, you are a deployer, and your own obligation is narrower: disclose deepfake-style content to the person seeing it, and disclose AI-generated text if you are publishing it to inform the public on a matter of public interest.

That distinction is why the honest answer to "am I compliant" usually starts with "is my vendor compliant," not with anything you personally changed in your sequences.

Where this touches outbound today

Most cold email and LinkedIn outreach does not trip either obligation in a meaningful way. Where it actually shows up:

• AI-generated personalized video, the Vidyard-style tools and AI-avatar SDR products that record a synthetic talking-head clip per prospect. If the avatar resembles an existing real person and could pass as authentic, deepfake disclosure applies to you as the deployer, and marking applies to the vendor as the provider.
• AI voice notes and voice clones used for LinkedIn or cold outbound. Same logic: if it could be mistaken for a real person's authentic recording, disclosure and marking both apply.
• Fully autonomous AI SDR agents that interact with a prospect directly, over chat or voice, without disclosing they are AI. The "you are talking to an AI" disclosure has applied since August 2 regardless of the marking grace period.
• AI-drafted email and LinkedIn copy is the one that worries people most and matters least here. Plain text sent one-to-one to a named prospect is not "informing the public on a matter of public interest," so the public-interest text-labeling duty almost never applies to a cold email. The marking duty on text output still technically sits with providers, but detection standards for text are the least mature part of the whole framework right now.

AI-generated outbound content, obligation by format

Step 1: inventory every AI-generated asset in your outbound

List every tool in your stack that generates audio, image, video or text rather than just sending it. Sequencing and sending platforms are not in scope on their own. A tool that clones a voice, animates an avatar, or writes copy from a prompt is. Most teams find this list is shorter than they feared, usually one or two tools, but they have never actually written it down.

Step 2: get each vendor's marking answer in writing

For every tool on that list, ask directly: does your output already carry a machine-readable mark, and if you were already on the market before August 2, 2026, what is your plan for the December 2 deadline? A vendor with a real compliance program will have a specific answer. A vague one is itself useful information.

This is the audit I now run before I let a client turn on an AI voice or avatar tool for prospecting: get the vendor's answer in writing before the campaign launches, not after a prospect or a regulator asks about it.

Step 3: build a deployer disclosure for anything deepfake-like

If a tool generates content that could pass as an authentic recording of a real person, your disclosure has to be clear and distinguishable, and it has to land at the first exposure, not buried in a footer or a terms page. For video, that means a visible label at the start of the clip. For voice, an audible statement before the message. Waiting until a prospect asks is too late under the rule.

Step 4: know what is actually exempt

Two carve-outs matter for outbound specifically. First, assistive editing tools that only correct grammar or tighten a sentence a human already wrote do not count as generating new content. Second, content that is evidently satirical, fictional or an obvious creative work has a lighter disclosure bar, though that exemption realistically never applies to a sales sequence. The one that actually protects most of you: one-to-one outbound copy is not "informing the public," so the text-labeling duty aimed at public-interest content almost never reaches a cold email.

Step 5: write it down

Keep a short record of which tools you checked, what each vendor told you, and the date you checked it. If a regulator or a prospect ever raises a question, a dated internal note showing you actually looked into this is worth far more than trying to reconstruct your due diligence after the fact.

Does this apply if you are not based in the EU?

Probably, if any of your outbound reaches people in the EU. The AI Act follows the same jurisdictional logic as GDPR: it applies based on where the AI system's output is used, not where your company is headquartered. Legal analyses of the Act's scope describe this as a market-location test rather than a strict targeting test, meaning a US or UK company whose AI-generated outreach reaches EU-based prospects can fall in scope even without deliberately marketing into the EU, per legal guidance on the Act's extraterritorial reach. If any meaningful share of your ICP sits in the EU, treat this as your problem too, not just a European competitor's.

What happens if you ignore it

Non-compliance with Article 50's transparency obligations can draw fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher, with proportionality considerations for smaller companies, per the European Commission's own FAQ. In practice, enforcement in the first months after a deadline like this tends to focus on egregious, visible cases rather than a B2B team that made a good-faith check and came up short on an edge case. That is a reason to actually do the check, not a reason to skip it.

FAQ