I've watched too many initiatives die twice: once when the team realized the 'AI' was a rules engine, and again when a regulator asked a question nobody could answer. The second death is new. Financial regulators have begun treating exaggerated AI claims as deceptive practice and, in some cases, securities fraud. That changes the math.
The lie wasn't in the demo. It was in the ownership.
AI washing happens because no single person owns the gap between what is claimed and what is shipped. Marketing wants a story. Product wants a launch. Engineering delivers a notebook. Compliance checks a box labeled 'human in the loop.' When the model drifts or the vendor changes terms, that chain has no owner—and no defined test for what AI actually is.
Incentives reward the label, not the capability
Why does this persist? Because internal budgets flow to 'AI' lines. Headcount gets approved for 'AI' roles. Fewer approvals come for 'a regression model that occasionally misfires.' The moment you name something AI, the economics of attention change. Most post-mortems find the technical team knew about the gap; they just weren't invited to the slide review.
Post-mortem: capability claims ran ahead of evidence
The failure pattern is consistent:
- The training data can't be traced, so nobody knows if the output is inference or lookup.
- There's no baseline comparison, so 'better' means 'we spent more.'
- There's no rollback, so turning it off becomes a customer-facing event.
Those are not marketing defects. They are engineering and leadership defects, dressed up as innovation.
The takeaway
Stop asking whether the legal disclosure is strong enough. Start asking whether the system can survive a simple audit: show the data lineage, the evaluation set, and the off switch. If you can't, it's not an AI initiative—it's a liability with a press release. The next enforcement action won't target the copywriter. It will target the executive who signed the roadmap.