Marking Your AI Replacement

Marking your AI Replacement

Moving into 2026, Cannes Lions made an interesting decision regarding AI.

After the AI scandals at Cannes Lions in 2025, as it set the rules for 2026, it could have capped the percentage of synthetic contribution to any work, or required a minimum of human authorship. It didn't. Instead, it made disclosure of AI use mandatory, with provable claims, requiring a named CEO or CMO to sign off, and the penalties for non-disclosure draconian. It put no limit on how much AI could be used.

The initial result was a rise to 40% of all entries disclosing AI use. Double the previous year, and up from 11% when disclosure was first introduced, as optional, in 2024. Asked about the figure, the Cannes chairman offered only that he didn't know whether 40% was a lot. The festival also reported an overall 25% reduction in entries, which it attributed to the improved governance.

By not policing authorship, Cannes has split the AI problem in two. Disclosure governs the input; it manages trust: is the entry what it claims to be? Human judgment governs the output; it manages worth: is it any good? Authenticate by process, and evaluate by person. This is a very clean way to let AI into a high-stakes system without surrendering standards: it keeps the judgment of worth in human hands.

But it has a hidden dependency nobody has answered. It only works if the human judge weighs the work on its merit, not on the AI disclosure. If the "AI was used here" quietly lowers the potential mark, disclosure stops managing trust and becomes bias. Cannes designed the rules to prevent the embarrassments of 2025, but how well does it handle the judges’ own stake in the importance of a human-dominated outcome?.

So the question that matters is straightforward: of the work that won, how much had disclosed AI? We don't know. (If you've found the data, I'd genuinely like to see it. I couldn't.) Cannes published that 40% of entries disclosed AI use, but nothing about how those entries fared. In the jury rooms, the signals were inconsistent: AI was described as "table stakes," no longer impressive on its own, even as jurors worried aloud about a two-tier system, human work judged in one room, machine work in another.

On one hand, AI is seemingly treated as neither positive nor negative, decided purely on merit. On the other, it may be bias in disguise, not that AI is worse, but that AI alone is unimpressive. A disclosure regime is only as trustworthy as the evenhandedness on the other side of it. Cannes built an elegant mechanism for surfacing the truth about AI's involvement going in, but told us nothing about whether that involvement is punished coming out.

Which leaves the real question unanswered, and it is sharper than just worthiness. The humans judging AI's output are, increasingly, the same humans whose work that AI output is competing with. We are asking the incumbent to rule fairly on the challenger, while the challenger is increasingly making a play for the incumbent's job. That is not a bias a disclosure form can remove; it is a conflict of interest in the room, and it will only deepen as the technology improves. Any organization folding AI into how it operates inherits the same structure: disclosure can manage trust on the way in, but worth is judged on the way out by people with a stake in the answer.