How to Commission AI-Assisted Video Safely
Updated September 17 2026
Commissioning AI-assisted video safely comes down to five things agreed before the first render. Which parts will be captured, generated or altered. Who owns or licenses each part, in writing. Which AI tools are used, what their terms allowed, and what happens to your material inside them. Who tells the audience, where, and in what words. And what records you receive at delivery. Put those five in the brief and the contract and most of the trouble never starts.
Agree these five before the first render.
I have commissioned, produced and delivered films for brands and charities for fifteen years, most of it filmed the traditional way. As AI has become part of more of the workflows I brief and sign off, I have watched it turn up inside real commissions, so I see it from the inside rather than from the outside. I write briefs, I sign contracts, and I am the one a client rings when a platform puts a label on their advert. This page is the checklist I use. It is not legal advice, and it does not pretend the rules are settled. It tells you what to ask for, what to write down, and what will happen to your video once it goes out.
This guide is for a brand, agency, charity or communications team commissioning AI-assisted video from a production company, freelancer or in-house team, and for the supplier delivering to them. It does not explain how to generate video, compare tools, or cover selling stock footage.
Picture a situation that is now common. An agency commissions a product film. The presenter and the product shots are real. The editor generates a clean studio backdrop and a two-second establishing shot of the city, and an AI tool tidies the presenter's audio. Nobody mentions any of this to the client, because it felt like ordinary post. The film goes on TikTok and Instagram and within the hour it carries an AI label on both, because the generator stamped the file with provenance data the platforms read. The client's first message is "why is our advert labelled AI?" The second is "do we own this?" And the contract says the client owns everything, which turns out not to be quite true of the two generated shots.
Every step was normal. Three things were never agreed. What would be generated, who would disclose it, and who owned the pieces. This page is about agreeing them first.
Explore this guide
If there is one section to start with, begin with What the contract must say. It is where most commissioning problems are either prevented or caused.
- The five things to agree before the first render
- What counts as AI use
- Do you need AI for this job?
- What to put in the brief
- What the contract must say
- What happens to your material inside the tools
- Approvals, revisions and reuse
- What the platforms may do to your video
- The rules that apply in the UK and the EU
- What you should receive at delivery
- Red flags, and what a good supplier looks like
- What to do when a platform flags it
- Questions people ask
- Where to go next
The five things to agree before the first render
If you only have ten minutes, write down the intended use, the platforms, the territories, what AI may and may not do, the ownership or licence outcome, who handles disclosure, and what you get at delivery. That is a usable brief.
| Before production | The question to answer | Where it lives |
|---|---|---|
| Scope | Which parts will be captured, generated or altered, and what is off limits? | The brief |
| Rights | Who owns or licenses each part, including the generated ones? | The contract |
| Tools and material | Which tools, under what terms, and what happens to our material inside them? | The contract |
| Disclosure | Who tells the audience, on which platform, in what words, and for paid as well as organic? | The brief and the contract |
| Records | What do we receive at delivery, and for how long is it kept? | The delivery pack |
What counts as AI use
The rest of this page uses three words, and they are the three words that should appear in every brief and every delivery schedule.
Captured means recorded with a camera or a microphone. The presenter, the product on the bench, the interview audio.
Generated means made by an AI system rather than recorded. A studio backdrop, an establishing shot, a synthetic voice, a scene that was never filmed.
Altered means real footage or audio materially changed by AI after capture. A cleaned voice track, a replaced sky, a face or a line that was changed.
Two more terms come up. Provenance is the information a file carries about how it was made or changed. Disclosure is what the audience or platform is told about AI use. Clean-up that changes nothing about what the video claims, such as noise reduction or colour correction, is assistive and needs no disclosure on any major platform. My AI policy sorts the same material into assisted, altered and synthetic, and the two vocabularies line up.
Do you need AI for this job?
AI is a tool, not a requirement, and the first thing a good brief does is decide whether it belongs in this job at all. It is often a strong fit for stylised sequences, backgrounds, concept work, variations of an approved shot and clean-up. It adds risk, approval friction and disclosure duties wherever the video is asking to be believed, a real person speaking, a product working, a place or an event that happened, and that test is set out in how audiences decide whether a video is real.
Three questions settle it. What is AI doing here that a conventional workflow would not do as well? Where is the risk, a real face, a real voice, a product claim, a brand cue, a sensitive setting? And is there a hybrid route, where AI does one narrow thing and the rest stays conventional? If the answers are not clear, keep AI optional. Sometimes the right answer is a crew and real light, and a good supplier will say so.
What to put in the brief
Most commissioning problems are about things nobody wrote down. The brand gets a file, then needs a new aspect ratio, a clean audio swap, a localised version or a shot removed, and if the only deliverable was an mp4, you are rebuilding under pressure. So the brief has two layers, what gets posted and what keeps the project maintainable afterwards. This is the commissioning spec I use, plain on purpose.
| Spec section | What to write | What to agree upfront | What to keep as proof |
|---|---|---|---|
| Goal and distribution | What the video must achieve and where it will run | Platforms, formats, captions, paid or organic, territories including whether it runs in the EU | Approved brief, channel list, approval message |
| AI use scope | Which parts will be captured, generated or altered, and what is not allowed | No-go areas such as a real person's likeness, voice cloning, brand lookalikes, sensitive themes | Scope note, restrictions list, reference pack version |
| Tools, terms and accounts | Which AI tools will be used and whose accounts | A dated copy of each tool's terms kept on file, and what the client needs from the account after delivery | Terms PDFs with dates, account details for handover |
| Client material | What of yours may be uploaded to each tool | Whether it is retained, whether it is used for training, who controls the account, and what must never be uploaded | Written list of permitted and prohibited uploads |
| Deliverables and handover | Everything delivered beyond the final export | Masters, platform versions, project files, generation log level, clean stems, licences and releases | Delivery manifest, version names, archive location |
| Approval chain | Who gives notes and who signs off | One decision-maker, checkpoints, what counts as sign-off | Approval messages, dated exports tied to approvals |
| Revisions and change control | How many rounds, and what becomes a change request | What a round is, how notes are consolidated, what moves cost or timeline | Change log, revision summaries, brief versions |
| Disclosure plan | Who discloses AI use, in what words, and where, for organic posts and for paid adverts | The wording, the placement on each platform, the ad-account setting, and who checks it before it goes live | The disclosure as posted, screenshots |
| Reuse rules | What the brand can reuse and what the supplier can reuse | Outputs, project files, stems, whether prompts and settings are reusable, limits by campaign | Reuse clause, handover receipt |
| Response plan | Who responds if a platform labels, limits or removes the video, and how fast | Poster of record, ad-account owner, supplier, the response window, who supplies the pack | Posted URLs, platform notices, timestamps |
Two rows did not exist in briefs two years ago. The disclosure plan matters because your video may well be labelled whether you plan for it or not, and deciding the wording in advance is the difference between a label you chose and a label that surprises your client. The client-material row matters because the question of what happens to your unreleased product, script or footage once it is uploaded to a tool is one most briefs have never asked.
Here is the same list as a checklist you can paste into a brief.
Before production starts, confirm: [ ] Intended audience, platforms and territories, including the EU [ ] Paid advert, organic post, or both [ ] Parts that may be captured, generated or altered [ ] Uses of AI that are not allowed [ ] AI tools, whose accounts, and a dated copy of each tool's terms [ ] What client material may go into each tool, and what never may [ ] Ownership or licence for each part, including generated parts [ ] Releases for every face, voice and performance [ ] Named approver and approval checkpoints [ ] Disclosure wording, placement and responsibility, per platform [ ] Delivery files, records and retention period [ ] Who responds if the video is labelled, limited or challenged
What the contract must say
Rights
Assignment or licence, spelt out, plus the schedule of what was generated.
Tools and material
Named tools, their terms on file, and what of yours may be uploaded.
People
A signed release for every real face, voice and performance.
Disclosure
Who discloses on each platform, named across the whole chain.
Records and reuse
The delivery pack, the retention period, and what each side can reuse.
Insurance
Whether the supplier's cover extends to generated and altered material.
Paying for a production does not, by itself, transfer every right in it. A finished film holds separate rights in its footage, script, music, performances, stock, graphics and any generated elements, and the supplier can only pass on what it holds. So the contract should state which rights are assigned or licensed to you, and for what territory, duration, media and purpose, and the supplier should have cleared its own contributors and third-party sources first. The full ownership picture, with sources, is on who owns an AI-generated video, and if the supplier or campaign is US-based, the client can own the film from the start as a work made for hire only where the signed agreement says so.
Generated material needs its own line, because ownership does not carry every issue. A generated clip used unchanged attracts no copyright in the US and an uncertain position in the UK, so there may be nothing to assign. The tool's terms separately govern what the output can be used for, and third-party rights in what went into the shot are not settled by ownership at all. So "the client owns everything" transfers everything the supplier has, and no more. The fix is a schedule listing what was captured, generated and altered, attached to the delivery. I put one in every agreement now. The template is on the who-owns page and a filled example is further down this one.
Then the clauses. Grouped so you can check them in a minute.
Rights. An assignment of copyright in the film to the client, or a licence with its scope spelt out, whichever is intended, not silence. The schedule of captured, generated and altered attached to the delivery. A warranty that the supplier holds releases for every real person and every voice, and licences for every piece of music, stock and font.
Tools and material. The AI tools named, whose accounts they run on, and a dated copy of each tool's terms kept on file, because the terms govern what the supplier was allowed to do with the output and they change. What client material may be uploaded to each tool and what never may. Whether the client needs anything from the supplier's account after delivery, to regenerate a shot, retrieve settings or continue the job elsewhere.
People. No real person's face, voice or performance is generated or altered without a signed release covering that specific use. Staff count as real people, and the employment relationship is one the ICO names as a power imbalance for consent, so their consent needs particular care. The face side is covered in AI consent and likeness in film production and the voice side in what I see from the side that hires voice actors.
Disclosure. Who is responsible for disclosing AI use on each platform, and the wording. Where three companies are in the chain, which is the common shape of a UK job, name all three. The poster of record, the owner of the ad account, and the supplier who hands over the schedule and the disclosure line. The EU's guidance puts the labelling duty on whoever controls the AI use rather than on whoever owns the campaign, and says that should be reflected in contracts, which is a good reason to write it down even for a UK-only job.
Records and reuse. The documentation pack the supplier delivers. How long it keeps the project files. Reuse rights for both sides, including whether prompts and settings are deliverables. Revision rounds, what counts as a round, and what triggers a change request. A response window for platform issues, and who responds first.
Insurance. One question to ask. Does the supplier's professional indemnity cover generated and altered material, including likeness? Many policies were written before this work existed.
If a supplier cannot explain which records they will provide, what they will keep, and what is excluded, ask for that distinction in writing before production starts.
What happens to your material inside the tools
When a supplier uploads your material to an AI tool, an unreleased product, a script, internal footage, customer data, brand assets, it leaves your control and enters that tool's terms, which decide whether the material is retained, whether it trains the model, and whether it can be deleted. Those terms differ by tool and by plan and they change. Business tiers often exclude training and add deletion rights; consumer tiers often do not.
So agree three things. What may be uploaded to each named tool and what never may, which for most brands rules out anything unreleased, anything with customer data, and anything under embargo. Whose account is used, because the account holder's plan sets the terms, and a freelancer's personal account may not offer what a business one does. And a dated copy of the terms that applied on the day, kept with the pack. I do not name specific tools' terms here because they move too fast to stay accurate.
Approvals, revisions and reuse
AI makes it easy to keep exploring when the job needs to converge, so treat approvals like version control. What I use has three parts. One round is one consolidated set of notes from one named approver. Notes that change the brief are change requests, not revisions. And approvals happen at checkpoints, the look, the rough cut, the fine cut, not only at final export, because changing an approved look after generation is expensive in a way it never was with a camera.
Reuse needs spelling out, because methods and outputs blur together in AI work. The clean split is that the client reuses the approved cuts, platform versions, project files and stems within the agreed scope, and the supplier reuses its general workflow where nothing confidential travels into another job. Prompts, settings and intermediate renders are agreed to the level of reproducibility the client is paying for, not delivered by default. If an agency briefs several creators against one asset pack, use one spec, one log level, per-cut approvals and one disclosure plan so the labels do not drift across uploads.
What the platforms may do to your video
A label can reach your video by three routes. The person uploading discloses it, the platform's systems detect it, or the file carries provenance data such as C2PA Content Credentials that the platform reads on upload. Those routes do not always agree, and provenance data is often stripped by an ordinary export, a transcode or a scheduling tool. So do not assume the label is in your control, or that the file will arrive with its provenance intact. Plan the wording in advance, and make sure the client sees the label before the audience does. Here is where each platform stands, checked on 17 September 2026 against the platforms' own announcements.
YouTube requires creators to disclose realistic content that was altered or generated, and its help page on altered or synthetic content sets out the duty. Since 27 May 2026 the label sits directly under the player on long-form video and as an overlay on Shorts, and if a creator does not disclose and YouTube detects significant photorealistic AI, it applies the label itself. Animated, stylised or slightly altered content keeps its disclosure in the expanded description. YouTube presents the label as information for viewers rather than a penalty, which is a different position from TikTok's removals and Meta's penalties for undisclosed content.
TikTok has required creators to label realistic AI content since September 2023, and since May 2024 it has read C2PA credentials on upload and applied its label automatically, the first platform to do so. Once TikTok applies that label, the creator cannot remove it. TikTok's guidance treats realistic AI-generated speech as in scope, including over real footage. Undisclosed realistic AI content can be removed or have its reach reduced.
Meta requires you to use its disclosure tool when you post organic content with photorealistic video or realistic-sounding audio that was digitally created or altered, and says it may apply penalties if you do not. It also reads industry provenance signals to label content automatically, across Facebook, Instagram and Threads.
Paid adverts are a separate duty, and the one a marketing lead is most likely to miss. TikTok's advertising policy, which is a different document from its community guidelines, requires a disclosure on realistic AI-generated ad creative, and an undisclosed advert can be rejected or restricted. The commercial-content setting is a separate toggle again. So the brief needs to record whether the cut will run as a paid advert, and the account that pays for the media, not only the editor who exported the file, has to know which disclosure tool to use.
| Platform | Plan for | Verify before posting |
|---|---|---|
| YouTube | Creator disclosure for realistic altered or generated content, and a platform-applied label if it is detected | The current help page wording and where the label will sit for your format |
| TikTok | Creator label for realistic AI content, automatic labelling from provenance data, and a separate advertising disclosure for paid creative | The current community guidelines for organic, and the advertising policies for paid |
| Meta | Disclosure tool for photorealistic video or realistic audio, and automatic labels from provenance signals | The current Facebook, Instagram and Threads requirements, organic and paid |
The disclosure lines I use, and the three tiers I sort material into, are in my AI policy.
The rules that apply in the UK and the EU
Regulatory position, checked 17 September 2026. This is a summary of two regimes, not legal advice, and both are worth checking before a major campaign is approved.
If the campaign falls within the EU AI Act's scope, its transparency provisions apply, and they have applied since 2 August 2026. Article 50 requires providers of generative AI systems to mark synthetic output in a machine-readable way, and requires deployers, which includes an organisation publishing AI-generated or manipulated content, to disclose it to the audience where the rules require. The Commission's FAQ and its final Guidelines of 20 July 2026 add four points a commissioner needs. The lighter treatment for artistic or creative work is read strictly, and all the advertising examples in the Guidelines require the standard label, so do not plan an advert around the exception. A video generated before 2 August 2026 does not need labelling retroactively, though the Commission encourages it. Generative systems already on the market have until 2 December 2026 to meet the machine-readable marking rule. And the Guidelines take a broad view of territory, so content posted on the open internet may be in scope if it is used in the EU, without specific targeting. Record the campaign's territories in the brief.
In the UK, the advertising rules apply however an advert was made. The Advertising Standards Authority's position, set out in May 2025, is that there is no blanket duty to disclose AI in an advert, the codes apply whatever the tool, and a disclosure is very unlikely to cure an advert that is misleading in the first place. So the question is not "did we use AI" but "does this advert give a truthful impression". A generated shot of a product working, presented as footage of it working, is what the codes catch. The ruling that illustrates it is on how audiences decide whether a video is real. Charities and public bodies carry an extra layer, fundraising claims, vulnerable audiences and public money, so check the relevant sector code, such as the Fundraising Regulator's, alongside the ASA.
| Area | What it means for the brief | The limit |
|---|---|---|
| UK advertising | Do not present generated or altered material as evidence of a real person, product or event if that impression would mislead | No blanket AI disclosure rule does not mean no risk |
| EU transparency | Record the territories, plan the label, and put the labelling responsibility in the contract | Advertising rarely qualifies for the artistic exception |
| UK copyright and contracts | Put assignment or licence terms in writing and clear contributors and third-party assets | Generated parts and underlying rights may need separate treatment |
Two things are still moving. The UK government has said it will consult on protections for a person's image and voice against unauthorised AI replicas, and a taskforce is working on labelling AI-generated content, with an interim report due this autumn. Neither had produced binding rules when this page was checked. Do not rely on the absence of a rule for a campaign with a long shelf life.
What you should receive at delivery
The maker's full proof trail is on who owns an AI-generated video. This is the commissioner's version, what should arrive with the final files, and what to ask for if it does not.
| What you get | Why you need it |
|---|---|
| The one-page schedule of captured, generated and altered, shot by shot | Answers "what is real" and "what do we own", and lets a single shot be swapped without re-briefing the film |
| The signed contract and any assignments | The main contractual evidence of the rights you acquired |
| A dated copy of each AI tool's terms, and the account they ran on | What the supplier was allowed to do with the output, on the day |
| Releases for every person and voice, licences for every asset | The inputs were cleared |
| The project file, masters and clean stems, with version names or checksums | You can reopen, fix and re-version without going back to the supplier |
| The disclosure wording as posted, organic and paid, with screenshots | What the audience was told, and where |
| Approval messages tied to dated exports | Who signed off what, and when |
| The posted URLs and any platform notices | Your record if a platform labels, limits or removes it |
| The agreed retention date for the project files | So you know when the ability to re-version expires |
Agree a retention period in the contract rather than assuming one. For a campaign that will be recut, anything with a real person in it, or any claim about a product, I would ask for longer than the supplier's default.
Here is what the schedule looks like filled in, using the product film from the top of this page. It is a labelled example, not a client's record.
| Shot | Status | Detail | Rights and disclosure |
|---|---|---|---|
| 01 to 09 | Captured | Presenter to camera, studio, two cameras | Presenter release on file, covers paid social and web for 24 months |
| 10 | Generated | Studio backdrop, [tool and version], replaced green screen | Tool terms PDF dated. No real person or brand in the generated element. Disclosed |
| 11 | Generated | City establishing shot, two seconds, [tool and version] | Tool terms PDF dated. Disclosed |
| 12 to 18 | Captured | Product on bench, macro | Product supplied by client |
| Audio | Altered | Presenter voice cleaned with [tool], no words changed | No disclosure required, clean-up only |
| Off limits | Not permitted | No face swap, no voice clone, no competitor lookalike | Restrictions list, brief v3 |
Red flags, and what a good supplier looks like
Five things that should make you stop and ask.
"We own everything" in the contract, with nothing about what everything includes or which parts were generated.
No named AI tools, and no record of the terms that applied.
No answer on face, voice or performance rights, or a release that does not cover the AI use.
Disclosure treated as the platform's problem, or nobody's.
Nothing delivered beyond the final mp4, and no answer on what is kept or for how long.
And the mirror. A well-run AI-assisted production looks like this.
AI use is scoped in the brief before anything is generated, with the off-limits list written down.
The tools are named, the accounts are known, and the terms are on file.
Every real person and voice has a release that covers the specific use, and every asset has a licence.
The disclosure wording and placement are agreed before delivery, for organic and paid.
You receive a schedule, the project files and the pack, and you know when the files expire.
What to do when a platform flags it
Most trouble arrives as an unexpected label, reduced reach, a takedown, or a message from someone who recognises themselves. A calm, quick response with the pack behind it settles most of it.
Capture the evidence first. Links, screenshots, timestamps, the exact version that went up.
Pause further uploads of that cut until the issue is understood.
Pull the pack for the approved version. Because the schedule is shot by shot, you can identify the one shot at issue rather than the whole film.
Respond through the platform's own process, in plain factual language, before anything goes public.
If one shot is the problem, swap the version. That is usually cleaner than winning an argument.
If a real person's face or voice is involved, or there is real commercial harm, get a lawyer in before you say more.
If the challenge is about a person rather than the platform, treat it as a consent question first. A label does not cure a missing release.
Questions people ask about commissioning AI-assisted video
What should an AI video brief include that a normal brief does not?
Five things. Which parts will be captured, generated or altered, and what is off limits. Which AI tools will be used, on whose accounts, with their terms kept on file. What of your material may go into those tools. A disclosure plan, the wording and placement on each platform, for paid as well as organic. And the records you expect at delivery.
Do I own the video if I commissioned and paid for it?
Not automatically. Payment does not transfer every right, and a film holds separate rights in its footage, script, music, performances and generated parts. In the UK the supplier is normally the first owner of what it made, and you have a licence for the agreed use unless the contract assigns the copyright to you. Put ownership in the contract, in writing, with a schedule of what was generated.
Can a production company use AI on my video without telling me?
Often, unless the contract says otherwise, which is why it should. Ask for a scope of AI use in the brief and a schedule of what was captured, generated and altered at delivery. A good supplier offers both without being asked.
Will my advert get labelled as AI on TikTok, YouTube or Instagram?
Quite possibly, if it contains a realistic generated or altered shot. TikTok has auto-labelled from provenance data since May 2024 and requires a separate disclosure on paid creative, YouTube applies its own label when it detects significant photorealistic AI, and Meta requires disclosure of photorealistic video or realistic audio. Labels can also be triggered by data in the file. Plan the wording in advance.
Do I have to disclose AI in a UK advert?
There is no blanket duty. The ASA's position is that the advertising codes apply however an advert was made, and that a disclosure will not fix a misleading impression. So the test is whether the advert is truthful. A generated shot presented as evidence of a product working is the kind of thing that fails it, label or no label.
Does the EU AI Act apply to my marketing video?
If the campaign is within the Act's scope, yes, since 2 August 2026, and the Commission's guidance takes a broad view of what counts as used in the EU. Providers must mark synthetic output, and organisations publishing AI-generated or manipulated content must disclose it where the rules require. Advertising rarely qualifies for the artistic exception. Record the territories in the brief and check the Commission's FAQ before relying on this summary.
What should the contract say about AI tools and my material?
Name the tools and the accounts. Keep a dated copy of each tool's terms and have the supplier warrant its use was within them. List what of your material may be uploaded and what never may, whether it is retained or used for training, and whether you need anything from the account after delivery. The tool's terms govern what the output can be used for, and they change.
Can I use a real employee's or customer's face or voice with AI?
Only with a signed release that covers that specific use, and for staff the employment relationship makes the consent question more delicate, not less. A disclosure label does not replace consent. Treat any real person's face, voice or performance as a release question first.
What records should I ask for at delivery?
The shot-by-shot schedule of captured, generated and altered. The signed contract and assignments. Dated copies of the tool terms. Every release and licence. The project file, masters and stems. The disclosure wording as posted. Approval messages tied to dated exports. The posted URLs. And the date the supplier's retention of the files ends.
Where to go next
Agree it before the first render
The safest commissions make the invisible parts of AI-assisted production visible enough to manage, without turning the job into admin. Decide whether AI belongs on the job, write down what will be captured, generated and altered, put rights, tools, your material, releases and disclosure in the contract, expect the label and choose its words, and ask for the pack at delivery. None of that slows a production down. It is what lets the finished video travel, get re-versioned, and survive a question long after everyone has moved on.
What changed in this version
This page replaces the version published in March 2026. Since then, YouTube moved its AI label under the player and began applying it automatically on 27 May 2026, TikTok's advertising policy made disclosure mandatory on realistic AI ad creative, and the EU AI Act's Article 50 transparency rules came into force on 2 August 2026 with the Commission's final Guidelines published on 20 July. My AI policy and the page on who owns an AI-generated video now carry the disclosure and ownership detail, so this page has been narrowed to the brief, the contract and the checklist. Checked on 17 September 2026. It will be updated when the UK labelling taskforce reports, when the digital replicas consultation opens, or if YouTube, TikTok or Meta change their disclosure rules.
Primary sources
YouTube. Disclosing use of altered or synthetic content, and YouTube's announcement of 27 May 2026 on label placement and automatic labelling
TikTok. Content Credentials and automatic labelling of AI-generated content, announced 9 May 2024. Creator labelling of realistic AI content required since September 2023. Advertising policies, TikTok Business Help Centre, on AI-generated ad creative
Meta. Disclosure requirement for photorealistic video and realistic-sounding audio that was digitally created or altered, announced February 2024
European Commission. FAQ on transparency obligations under Article 50 of the AI Act, applying from 2 August 2026, and the Commission Guidelines on Article 50 of 20 July 2026
Advertising Standards Authority. Disclosure of AI in advertising, May 2025
UK government. Report on Copyright and Artificial Intelligence, 18 March 2026, including the digital replicas and labelling commitments
Information Commissioner's Office, guidance on consent and the employer and employee relationship
