By Sagar Shankaran, Founder of CallSphere
The EU AI Act's August 2, 2026 date reaches US studios delivering to EU viewers. Here is what to log, disclose and consent for, and what is out of scope.
Key takeaways
Short answer: if you delivered a cut that runs in front of viewers in the EU, and something in that cut was generated or altered by AI, yes — parts of it reach you. The Act's high-risk and transparency obligations carry a compliance date of August 2, 2026, and its reach is deliberately extraterritorial: it applies based on where the output lands, not where your suite is.
That sentence has been repeated at every trade event this spring, usually followed by something vague about governance. So here is the specific version for a production company. Almost none of the Act's heavy machinery — the risk classifications, the conformity paperwork, the technical documentation that runs to hundreds of pages — is aimed at you. One part is: the transparency rule that says when video content has been artificially generated or manipulated in a way that resembles real people, places or events, viewers have to be told, and the content has to be marked in a way machines can detect.
If you have replaced a sky, cloned a client CEO's voice for a German version, aged up a product shot from stills, generated background crowd, or upscaled 2011 archival footage of a real factory floor, you are in the room where that rule lives.
Strip out the legal language and the operational burden comes down to three artifacts, and none of them is new software.
One: a per-deliverable AI-use record. Which tool, which shot, which timecode range, what it did. This is a spreadsheet column in the same delivery sheet you already send with masters. Your assistant editor fills it in as they work; if it is filled in afterwards from memory, it will be wrong.
Two: viewer-facing disclosure where the content depicts something real. A generated sunset behind an entirely fictional product does not need a card. A synthetic voice speaking as a named executive does. In practice this is either a super, a line in the end card, or a disclosure in the caption and metadata of the delivered file, and the agency decides the placement — but you have to raise it.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent in your browser — 60 seconds, no signup.
Three: signed consent for any digital replica of a person. Talent releases written before 2024 do not cover a voice clone or a digital double. If you shoot SAG-AFTRA commercials, the union's digital replica consent provisions already require specific, separate consent, and California's digital-replica consent laws have been in force since the start of 2025. Your standard release needs a rider, and the rider needs to be in the delivery folder, not in someone's inbox.
flowchart TD
A["Editor drops a generated element into the v6 cut"] --> B["Log tool, shot and timecode in the AI-use sheet"]
B --> C{"Does it depict a real person, place or event?"}
C -->|"No"| D["No viewer disclosure; keep the log"]
C -->|"Yes"| E["Agree disclosure wording with the agency"]
E --> F{"Is a person's likeness or voice replicated?"}
F -->|"Yes"| G["Pull the signed replica consent rider"]
F -->|"No"| H["Attach Content Credentials to the master"]
G --> H
H --> I["Deliver to the client asset library with the log"]
This is the part nobody tells you, and it saves you weeks. You are a user of these tools, not the company that built them. The obligations that fall on the builder — the conformity assessments, the model documentation, the registration duties — belong to Adobe, OpenAI, Google, Runway and the rest, not to you. You do not have to certify anything about how a generative tool works internally, and you cannot; you did not build it.
California's SB 53, which took effect January 1, 2026, is aimed squarely at large frontier AI developers — the companies training the biggest models. A production company is not one. Texas TRAIGA, effective the same day, is broad in who it touches but is fundamentally about intentional harm, prohibited manipulative uses and government use of AI; a studio that discloses its synthetic content and gets consent for replicas is not the target. Colorado, New York, Utah, Nevada, Maine and Illinois each have their own statutes, which matters if you have staff or clients in those states, and federal preemption of state AI rules is still unsettled as of this month — so do not assume Washington is about to clear the board for you.
What you should assume is that the requirement reaches you commercially before it reaches you legally. Agency procurement teams have started putting AI-use declarations into deliverable specs and vendor questionnaires. The studio that answers that question in ten minutes with a real log looks like a professional operation. The one that says "let me ask the editor" does not.
Your post supervisor is cutting a master, four cutdowns, six socials and two localised versions — one German, one Spanish — for a campaign flighting in the US, Canada, Germany and the Netherlands. Under the old habit, that folder goes out with a delivery sheet listing codec, frame rate, aspect ratio, loudness and slate information, and nothing else.
Under the new habit, it goes out with one extra tab. Row by row: the German version uses a synthetic voice of the same non-union VO artist who recorded English, with a signed replica rider dated April 14. Shot 12 in the master has a generated sky replacement, no real place depicted, no disclosure required. Shots 3 and 4 use upscaled 2011 archival of the client's actual Ohio plant, and the plant is a real place, so the end card carries the disclosure the agency approved on Tuesday. Content Credentials are attached to the delivered files out of Premiere Pro. Total additional effort on this delivery: about twenty-five minutes, most of it spent finding the rider.
| Assumption | Value |
|---|---|
| Delivered masters, versions and cutdowns per year | 140 |
| Share containing a generated or manipulated element | 35% (49 files) |
| Logging and disclosure time per affected file | 12 minutes |
| Loaded cost of assistant editor time | $60 per hour |
49 files × 12 minutes = 9.8 hours a year, or roughly $590. That is the entire ongoing cost, once the rider is in your release template and the column is in your delivery sheet.
Now the other side. Suppose one campaign gets pulled from flight in Germany because the localised version used a cloned voice and nobody can produce consent paperwork. You re-version six deliverables, the agency bills its own legal review, the media buy shifts, and your master service agreement almost certainly contains an indemnity clause pointing at you. Illustratively, call the direct cost $45,000 — and the real cost is that you are not on the next bid list. The point is not that this will happen to you. The point is that $590 a year is cheap insurance against something your MSA already makes your problem.
Still reading? Stop comparing — try CallSphere live.
CallSphere ships complete AI voice agents per industry — 14 tools for healthcare, 10 agents for real estate, 4 specialists for salons. See how it actually handles a call before you book a demo.
Three decisions should never be made off a template. First, whether a specific manipulated shot depicts something "real" — a de-aged founder, a composited building that resembles a real headquarters, a recreated news moment. That is a judgement call and it belongs to the agency's counsel with your facts in front of them. Second, the wording and placement of any on-screen disclosure, because that is a brand decision as much as a legal one. Third, anything involving a deceased person's likeness, which has its own rules in California and elsewhere and is not a place to improvise.
Also: do not use an AI tool to decide your compliance status. Use it to assemble the record — read the MSA, pull the indemnity and IP clauses, cross-check them against your delivery log, and hand the summary to a human who signs their name to the answer.
You need the consent riders and the internal log regardless, because state law and union agreements already push that way and because your clients will ask. The EU disclosure piece only bites when a deliverable reaches EU viewers — but check before you assume it does not. A global brand's "US campaign" ends up on a European site more often than the brief admits.
Ordinary post work that cleans or corrects what the camera captured is not the target. The rule is about content that is generated or manipulated so it resembles real people, places or events. Denoising an interview is not that. Putting words in the interviewee's mouth is.
You supply the facts for what happened in your suite; the agency and the advertiser own what the finished piece says to the public. Put that split in writing in your SOW, because otherwise it defaults to whoever gets asked first.
That is the most likely way this goes wrong. Add one line to your freelance deal memo requiring disclosure of any generative tool used on delivered elements, and ask on the day, not at delivery.
The smallest useful first step this week: add a column to your delivery sheet, add a digital-replica rider to your standard talent release, and add one line to your freelance deal memo. Everything else can wait until an agency asks.
One place this touches the phone: when a client or an agency producer calls asking what was used on a spot, the question usually arrives while your post supervisor is in a session. CallSphere builds AI voice and chat agents that answer your studio line and web chat, capture exactly what was asked and which campaign it concerns, and route it to the right person instead of leaving it in voicemail. Those agents also identify themselves as AI to the caller, which is one small disclosure box you get to tick without thinking about it.
Written by
Sagar Shankaran· Founder, CallSphere
LinkedInSagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
The EU AI Act's August 2 date, Texas TRAIGA and California SB 53 all landed. What a US rehab clinic must document, disclose and log, and what it can skip.
Whether a spot can run again depends on the MSA, talent releases, sync licence and change orders. In 2026 the entire file fits into a single question.
Driver screening tools and cab-facing cameras put carriers under Illinois, Texas and NYC rules in 2026. What to document, and what is genuinely out of scope.
Two casino systems land in the EU AI Act high-risk bucket: face matching and marker scoring. What US gaming operators document, disclose and can ignore.
What a US registered investment adviser must actually document, disclose and log in 2026 — and which AI statutes are genuinely out of scope for the firm.
Admissions, placement tests and proctoring put private schools inside 2026 AI rules. The inventory, the file, the parent notice, and what is out of scope.
© 2026 CallSphere Inc. All rights reserved.
Made within San Francisco
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.
Try Live DemoBook a DemoCalculate Your ROI