AD FIELD NOTESAI advertising · the work behind the output

Case files / generative experiential

Fishy.AI: make governance visible before it becomes a policy slide

IBM and Adobe used Firefly-made surreal fish across the Las Vegas Sphere, summit environment and a mobile experience to make AI trust and governance physically legible.

Campaign period: 2025Source published: Date not statedArchive entry prepared: 15 September 2026

Retrospective coverage, prepared in September 2026 from the original record.

The Las Vegas Sphere is turned into a glowing blue fishbowl full of surreal fish.
Fishy.AI campaign imagery on the Las Vegas Sphere. IBM / Adobe / Ogilvy; image published by WPP.

Reported execution: the fish escaped the deck

WPP describes Fishy.AI as an Adobe Summit activation for IBM and Adobe. Adobe Firefly generated surreal fish for a Sphere takeover, IBM’s booth, out-of-home placements, keynote material and a mobile-led experience. Attendees could ‘catch’ fish and receive an education point about unreliable or poorly governed AI output.

The case says the 90-second Sphere activation appeared every 30 minutes across the four-day summit. Its message, ‘Trust What You Create,’ linked the comic images to IBM watsonx and Adobe cloud tools. This was not a demonstration that an image model has solved governance; it was a campaign that made the governance problem into a public visual event.

Our reading: an abstraction needs a memorable consequence

Hallucination, bias and auditability are usually described in language that does not travel far beyond a conference session. The fishbowl gives each concern a shape without pretending the image itself is evidence. A strange fish is a useful warning sign because it is instantly legible as something that does not belong.

The metaphor also benefits from being deliberately silly. A solemn visual treatment could have repeated the technical tone the work was trying to interrupt. Humour gets attention; the surrounding game and education points give that attention somewhere to go.

Production lesson: keep the metaphor tied to a next action

For an experiential brief, decide which claim the image is allowed to make and which has to be explained elsewhere. A visually broken output can introduce a conversation about trust, but cannot demonstrate model controls on its own. Make the next interaction specific: a short lesson, a product explanation or an invitation to inspect a workflow.

The asset system should also hold together across a building-sized screen, a phone and tabletop material. Version prompts and final selections, and test whether the joke remains readable at each distance. Consistency is not only a brand requirement here; it keeps the governance message from dissolving into spectacle.

Evidence boundary: event metrics are not a governance audit

WPP reports funnel, understanding, trust, earned-impression and social-engagement figures. They are useful as agency-reported campaign outcomes, not as proof that attendees adopted safer AI practices or that any platform is inherently trustworthy.

A stronger follow-up would measure whether people could identify a concrete control after interacting with the activation, and whether that knowledge changed a later decision. The durable archive lesson is modest: generative imagery can dramatize a risk when the campaign is candid about the difference between awareness and assurance.

Sources & limits

WPP agency case study; event mechanism is documented by the maker, while reported outcomes are campaign claims.

  1. WPP — Ogilvy: IBM and Adobe’s Fishy.AI — source publication date not stated.

Send a correction with the passage and supporting source.

The weekly field note

One useful note for the next brief.

Campaigns worth studying, the work behind them, and questions to bring into your next creative review. Delivered weekly.

How we handle your email