What AI washing is
AI washing is marketing that claims more artificial intelligence than the product actually contains. Sometimes it is a rules engine dressed up as machine learning. Sometimes it is a "fully automated" workflow that a person is quietly running in the background. For a mid-market buyer it matters because the exposure has shifted. Regulators used to chase companies that misled consumers. Increasingly, they are chasing companies that misled business buyers, which is you.
The enforcement arc
The pattern is worth watching because it is escalating. In March 2024 the SEC settled its first AI-washing cases against two investment advisers, a modest US$400,000 in total[1]. By January 2025 it had charged its first public company. By April 2025 it had gone criminal: the founder of the "AI shopping app" Nate was charged with fraud after the app’s supposedly automated purchases turned out to be handled almost entirely by contract workers[2]. The FTC, through Operation AI Comply, took down the "world’s first robot lawyer" and, in a separate case, an AI content detector advertised as 98% accurate that independent testing put closer to 53%[3].
The through-line is simple. What is being punished is the gap between the demo and the delivery. When you are the one who paid for the demo, that gap is your loss first and the regulator’s case second.
The Australian picture
The same pattern is playing out in Australia. In October 2025 the ACCC took Microsoft to the Federal Court over how it presented Copilot-integrated Microsoft 365 plans to roughly 2.7 million Australian subscribers, alleging it concealed cheaper Copilot-free options[5]. The penalties behind the Australian Consumer Law also got sharper: from 28 March 2026 the maximum rose to $100 million per contravention. Then-ASIC chair Joe Longo had already warned in 2024 that ASIC was "on the lookout" for AI washing, and its REP 798 review documented how far governance was lagging adoption.
One honest caveat: neither ASIC’s nor the ACCC’s formal 2026 priorities name AI washing as a standalone item. The exposure runs through general misleading-conduct law, not a bespoke AI offence. For a board, that is the point. You cannot wait for a specific rule, because the existing rule already applies.
Agent washing is the 2026 version
The current wave of overselling has a name. Gartner calls it "agent washing": rebranding a chatbot, a script, or a bit of robotic process automation as an autonomous "agent". Gartner estimated that of the thousands of vendors claiming agentic capability, only around 130 genuinely have it, and predicted more than 40% of agentic AI projects will be cancelled by the end of 2027[4]. When a vendor promises an agent that runs your service desk end to end, the base rate says be sceptical.
A buyer’s test, taken from the case files
Every enforcement action above is also a question you can ask before signing. The FTC’s own guidance to sellers reads, inverted, as a buyer’s checklist[6].
- Who or what actually does the work? Ask for the share of tasks completed with no human in the loop, and get it in writing (Nate’s was near zero; Presto’s over 70% needed a person)
- Who owns the model? A vendor reselling someone else’s AI as its own is a documented pattern, not a hypothetical
- Where is the independent accuracy evidence? "98% accurate" tested at 53% is why you ask for third-party validation, not a slide
- Run a proof of concept on your own data, with a measurable success metric agreed before you start
- Re-test after deployment. A claim true on 1 January is not automatically true on 1 June, and models drift