Runway and the EU AI Act: A Compliance Guide for Generative AI Video and VFX

Probably not high-risk. Runway generates and manipulates video — text-to-video scenes, image-to-video shots, and video-to-video restyling of real footage — so the defining EU AI Act question for your organisation is Article 50 transparency under Regulation (EU) 2024/1689, not the high-risk regime.

The realistic duties are narrower than the headlines: Article 50 transparency where output reaches an audience, Article 4 AI literacy, and likeness, consent and IP governance for the people and footage you feed in. This is the generative-creative video and VFX playbook — distinct from our Synthesia guide and the full Article 50 deepfake labelling rules.

---

Where Runway sits in the EU AI Act: transparency, not high-risk

Most creative video generation is not high-risk. High-risk classification under Article 6(2) arises only where a system is used for a purpose listed in Annex III — point 4(a) recruitment, point 5(b) creditworthiness. Generating a film clip, an ad spot, or a VFX shot is not on that list. Runway is also not product-embedded under Annex I, so the Article 8-15 high-risk stack does not apply by default; the obligation is the limited-risk transparency tier in Chapter IV.

The responsibility split is the spine of this guide. Runway, the generator, carries the Article 50(2) machine-readable marking duty; your company, the deployer that publishes the output, carries the Article 50(4) deepfake disclosure duty. The two stack, and neither discharges the other.

---

Generation vs. manipulation: why most Runway output is a deepfake under Article 3(60)

Article 3(60) defines a deep fake as AI-generated or AI-manipulated image, audio or video content resembling existing persons, objects, places, entities or events that would falsely appear authentic. The test is the appearance of authenticity — not whether a named real person is depicted.

Runway's distinctive surface is two-fold: text-to-video and image-to-video generation of wholly synthetic scenes, and video-to-video manipulation of existing real footage — style transfer, object alteration, inpainting. Both fall in scope wherever the result reads as authentic, and photorealistic b-roll typically meets Article 3(60).

Clearly stylised, abstract, or evidently non-realistic output may fall short of the Article 50(4) disclosure trigger. But even there the Article 50(2) machine-readable mark still applies: the lighter case removes the human-facing disclosure, not the provider's mark.

---

The two duties: provider marking (Art 50(2)) and deployer disclosure (Art 50(4))

Runway's machine-readable marking duty — Article 50(2)

Article 50(2) requires the provider of a generative system to mark synthetic image, video, audio or text output in a machine-readable format, detectable as artificially generated, using a solution that is effective, interoperable, robust and reliable as far as technically feasible. Interoperability is load-bearing: a proprietary watermark only Runway's own tooling can read does not satisfy the duty. In practice this points towards open standards such as C2PA.

Your disclosure duty as the deployer who publishes — Article 50(4)

Article 50(4) requires the deployer that generates or manipulates a deepfake to make a clear, human-perceptible disclosure that the content is artificially generated, at the latest at first exposure under Article 50(5). A machine-readable mark a viewer cannot perceive does not satisfy this; Runway embedding a mark does not relieve your team of the on-screen disclosure, and vice versa.

When standard editing is excluded from marking — the Article 50(2) assistive-editing carve-out

The Article 50(2) marking duty does not apply where the system performs an assistive function for standard editing or does not substantially alter the input data — minor colour correction or cleanup. But Runway generating a scene, extending a shot, replacing an object, or synthesising motion is well past standard editing and must be marked. The burden is yours to show an edit was assistive.

The lighter disclosure for evidently creative work — Article 50(4)

The creative-work carve-out is acutely relevant to film, ad and agency use. Where a deepfake forms part of an evidently artistic, creative, satirical or fictional work, the Article 50(4) disclosure is reduced to a non-prominent form — but not eliminated, and a realistic AI endorsement in a commercial advertisement is not 'evidently artistic'. Separately, the Article 25 role-flip: a company that white-labels or substantially modifies a generative video capability and ships it under its own name can become a provider and inherit the Article 50(2) marking duty.

---

Likeness, consent and training-data / IP exposure

Generating or manipulating recognisable real people — actors, public figures, members of the public caught in source footage — engages likeness and consent before any AI Act question. You need a GDPR lawful basis under GDPR Article 6, and special-category protections under GDPR Article 9 where facial data is processed. The transparency layer sits on top of GDPR and IP law; a disclosure does not cure an unconsented likeness or an infringing output.

Training-data and copyright exposure is a Runway-distinctive concern: generative video models raise IP questions around training corpora and output similarity. Treat this as a contractual and IP-diligence matter — model terms, indemnities, usage rights — flagged for legal review, not an Article 50 obligation. Keep a governance record of source-footage rights, likeness consents, and the terms governing output ownership.

---

Worked example: a 250-person creative agency using Runway for client video

Lumen Atelier, a 250-person EU creative and advertising agency, uses Runway for text-to-video b-roll, AI VFX shot extension, and video-to-video restyling of client footage. Runway is the provider (Article 50(2) marking). Lumen Atelier is the deployer publishing client deliverables, carrying the Article 50(4) disclosure duty wherever the output reaches an audience.

Five concrete actions cover Lumen's position:

• Confirm Runway's marking and C2PA provenance capability for every output.
• Add a clear, human-perceptible disclosure to deepfake-grade deliverables.
• Apply the lighter, non-prominent disclosure to evidently creative pieces, but full disclosure to realistic ads.
• Secure source-footage rights and likeness consents with a defined scope.
• Deliver Article 4 AI literacy and log each project in an inventory register.

These deliverables are not high-risk. Had Lumen instead used Runway output to drive an Annex III decision — a recruitment or creditworthiness outcome — the high-risk regime would apply, triggered by that use.

---

Deadlines, penalties and a 2026 action plan

Under the Digital Omnibus, the Article 50 content-marking and watermarking obligations move to a fixed calendar date of 2 December 2026; the standards-contingent 'stop the clock' variant was rejected, so it is a hard date. Freshness note: the Digital Omnibus was adopted by the European Parliament on 16 June 2026 and the Council on 29 June 2026, and enters into force on Official Journal publication, expected before 2 August 2026 — so the 2 December 2026 content-marking date is settled.

Already in force and unaffected: Article 5 prohibitions and Article 4 AI literacy since 2 February 2025; GPAI obligations under Articles 51-55 since 2 August 2025, on a separate track.

Breach of the Article 50 transparency duties falls under Article 99(4): up to €15 million or 3% of total worldwide annual turnover, whichever is higher.

The €35 million or 7% top tier under Article 99(3) is reserved for the Article 5 prohibited practices and does not apply to labelling failures; the third tier is 1%, for incorrect information. None of this waits on the high-risk track.

---

How Confir helps

Register Runway in your AI inventory and Confir's deterministic, rule-based classification engine runs a transparency-trigger checklist from how you describe the system — whether output is realistic enough to be a deepfake, whether it is generation or assistive editing, and whether you hold the provider or deployer role. The workflow captures the Article 50(2) marking and Article 50(4) disclosure duties, records the footage-rights and likeness-consent position per project, and produces an auditable record.

The synthesis engine is deterministic and rule-based — no model inference, no hallucination. The same system description yields the same obligations every time. General-purpose AI workflow support remains on the roadmap, not marketed as complete.

---

Frequently asked questions

Is Runway high-risk under the EU AI Act?

Not in ordinary creative use. Generating b-roll, VFX shots, or restyled video is not an Annex III purpose under Article 6(2), so Runway output is not high-risk; it sits in the limited-risk transparency tier. The real duties are Article 50 transparency where the output reaches an audience, Article 4 AI literacy, plus likeness, consent and IP governance. High-risk would only arise from the use — feeding output into an Annex III recruitment or creditworthiness decision — not from the video-generation step, which is assessed separately.

Do we have to label every video we make with Runway?

Usually yes for realistic output. A photorealistic generated scene or AI-manipulated footage of a real place or event typically meets the Article 3(60) deepfake definition because it would appear authentic, so as the deployer publishing it you owe a clear, human-perceptible disclosure under Article 50(4), at the latest at first exposure. Clearly stylised or evidently non-realistic output may fall short of that trigger, but all synthetic output still needs the Article 50(2) machine-readable mark embedded by the generative system.

Does the 'standard editing' exception cover Runway VFX work?

Only the lightest edits. Article 50(2) says the marking duty does not apply where the system performs an assistive function for standard editing or does not substantially alter the input data — minor colour correction or cleanup. Runway generating a scene, extending a shot, replacing an object, or restyling footage substantially alters the input and is well past standard editing, so the marking duty applies. Treat the exclusion narrowly; the burden sits with you to show an edit was genuinely assistive.

How does the creative-work carve-out apply to ads and films made with Runway?

It lightens the disclosure but does not remove it. Under Article 50(4), where a deepfake forms part of an evidently artistic, creative, satirical or fictional work, the disclosure becomes non-prominent so it does not hamper enjoyment. That helps clearly creative film and stylised content. But a realistic AI-generated endorsement or product scene in a commercial advertisement is not evidently artistic, so it carries the full, clear disclosure. If a reasonable viewer could take the content for genuine, the lighter regime does not apply.

Who marks Runway output — Runway or our company?

Both, at different points. Runway, as the provider of the generative system, carries the Article 50(2) duty to embed a machine-readable mark in every synthetic output, using a solution that is effective, interoperable, robust and reliable — pointing towards open standards such as C2PA. Your company, as the deployer that publishes the video, carries the Article 50(4) duty to make a clear, human-perceptible disclosure. The provider's mark does not discharge your disclosure, and vice versa.

What about training data, copyright and depicting real people?

These sit alongside, and before, the Article 50 transparency question. Depicting or manipulating recognisable people engages a GDPR lawful basis under GDPR Article 6 and special-category protections under GDPR Article 9 for facial data, so secure documented likeness consent with a defined scope. Training-data and output-similarity copyright exposure is an IP and contractual matter — check the provider's terms, usage rights, and indemnities, and flag it for legal review. The disclosure does not cure an unconsented likeness or an infringing output.

When do the Runway content-marking rules apply and what are the penalties?

Under the Digital Omnibus the Article 50 content-marking obligations move to a fixed date of 2 December 2026; the 'stop the clock' variant was rejected. This is adopted under the Digital Omnibus — the European Parliament voted on 16 June 2026 and the Council adopted it on 29 June 2026 — and enters into force on Official Journal publication, expected before 2 August 2026. Breach of Article 50 falls under Article 99(4): up to €15 million or 3% of worldwide turnover. The €35 million or 7% tier (Art 99(3)) is reserved for Article 5 prohibitions, and SMEs benefit from the Article 99(6) cap at the lower amount.

---

Related guides

• the full Article 50 deepfake labelling rules
• Article 50 transparency obligations
• the avatar-video angle in our Synthesia guide
• the limited-risk transparency tier
• add Runway to your AI inventory