Eros Innovation has launched a UK innovation challenge on “trusted cultural AI”, with support from Innovate UK Business Connect. It is seeking tools for rights, content provenance, identity and consent across film, characters and synthetic media. Its subject is real and now regulated. The sponsor’s own numbers deserve a closer look.
What was announced
The challenge, “Trusted Cultural AI Infrastructure: Film, Character & Rights”, is looking for UK innovators building machine-readable rights and automated licensing, provenance tracking across the content lifecycle, and identity and consent management.
Selected solutions may progress to a proof of concept of about three months, with potential support of £25,000 to £50,000 each. Eros is seeking up to four solutions. Applications close on 13 November 2026.
The framing is a genuine question. As generative AI moves through the creative industries, rights, provenance and consent have to become part of the infrastructure rather than an afterthought.
The claim, and its weak point
Eros says its cultural data assets have received “independent valuations totaling approximately US$3 billion, including a valuation by OxValue.AI… alongside other independent valuation work”.
Its own earlier releases tell a narrower story. They state the dataset is “independently valued at USD 1.7 billion by OxValue.AI”. The figure has grown to $3 billion, with no second valuer named.
There is a further problem. OxValue.AI is not a neutral party. Eros Innovation and OxValue.AI announced a strategic joint venture in February 2025 to transform technology valuation in India. The valuer and the valued are business partners.
Eros’s materials are also inconsistent on the underlying assets. One figure is stable: 1.5 trillion tokens. Its library is variously 11,000 films, “more than 11,000 films” and “12,000+ films”, and the heritage is “four decades” in one place and “five decades” in another.
Why content provenance is now infrastructure
The reason this matters beyond one company is that content provenance has moved from a nice-to-have to a legal requirement.
The EU AI Act’s Article 50 obligations took effect on 2 August 2026. They require AI-generated and AI-manipulated content distributed to EU audiences to be marked in a machine-readable format, with penalties reaching €15 million or 3% of global turnover.
The EU’s Code of Practice names C2PA Content Credentials as the recommended mechanism. C2PA is an open standard that signs tamper-evident provenance metadata into a file. Its Content Authenticity Initiative now has more than 6,000 members.
The standard has also grown a consent layer. The CAWG consent assertion lets an asset carry a machine-readable pointer to an authoritative source of consent, covering AI training, licensing posture and compensation. That is the plumbing the challenge is asking for.
The licensing market is arriving
The commercial backdrop is moving just as fast. More than 140 lawsuits are pending over whether training on copyrighted works without consent is fair use.
A US federal appeals court has upheld a ruling that copying material to build training data was not fair use, in Thomson Reuters v ROSS. Bartz v Anthropic held that retaining pirated copies is infringing, and led to a reported $1.5 billion settlement.
Licensing is becoming the settlement mechanism. Universal Music Group and Warner Music Group struck deals with Stability AI and Udio. Suno and Udio have both signed deals after being sued.
The pressure is measurable. The RIAA and NMPA cited Deezer’s disclosure that it receives more than 50,000 fully AI-generated tracks a day, over 34% of its daily deliveries. A CISAC study projects AI-generated music will be worth a cumulative $42 billion by 2028, with a 24% loss of creators’ revenue.
What “rights-cleared” is worth
This is the context in which Eros’s positioning makes sense. Owning a large, rights-cleared corpus is a genuine differentiator when everyone else is being sued over scraped data.
But the value of that corpus depends on the rights actually being clean, and on the valuation being defensible. A joint-venture partner’s estimate, restated upward without a second source, is not the same as an audited asset.
The UK government has consulted on a text-and-data-mining exception with rights reservation, and the European Parliament voted in March 2026 against flat-rate global licensing for AI training. Both point the same way: rights holders want to be paid, and provenance is how payment gets tracked.
What to watch
Whether the challenge produces interoperable tooling or a set of bespoke proofs of concept. Whether C2PA and the consent assertion become the default, or fragment across competing standards. And whether the valuations behind “rights-cleared” datasets are ever independently audited, because the licensing market’s credibility depends on it.
The question is a good one. The answer will be settled by audits, not announcements.
Editor’s Note
This analysis draws on the Eros Innovation and Innovate UK Business Connect challenge release, Eros Innovation’s published materials, OxValue.AI and Eros Innovation’s joint-venture announcement, C2PA and CAWG specifications, the EU AI Act Article 50 and the EU Code of Practice on AI-Generated Content, US and UK copyright litigation and consultation material, and licensing and market data reported by the RIAA, NMPA and CISAC. Where a figure is a company’s own or an estimate, we say so.