AI Compliance & Regulation
Agent Washing: Telling a Real AI Agent From a Rebrand
"Agent washing" is when a vendor rebrands a chatbot as an "AI agent". Here's how Australian SMEs test the claim before paying — and the law that protects you.
The "AI agent" in the sales deck might be a chatbot in costume
Every software vendor is selling you an "AI agent" this year. Most of what they're selling isn't one. Gartner has a name for the gap — "agent washing", the rebranding of ordinary chatbots, rule-based automations and RPA scripts as autonomous agents — and it forecasts that more than 40 per cent of agentic AI projects will be scrapped by the end of 2027, undone by runaway cost, unclear value and weak controls.
For an Australian business writing the cheque, the risk was never the technology. It's paying agent prices for chatbot capability — then wearing the switching cost, the wasted staff hours and the customers served badly while you work out the tool can't do what the demo promised.
What "agent washing" actually is
An AI agent is not a chatbot. A chatbot answers; an agent acts. To act, it has to be wired into your real systems — your inbox, your job-management software, your accounting file — and handed authority to take steps on its own, in a loop, until the task is done. That wiring is where the capability and the risk both live, as Australia's cyber agency spelled out in its guidance on agentic AI. Agent washing is slapping the "agent" label on something that still needs a human to start and steer every step.
Gartner coined the term in June 2025 and estimated that only a small fraction of the thousands of self-described agentic vendors offer genuinely autonomous capability. The 40 per cent cancellation figure is a forecast, not a tally — revisited by Forbes in July 2026 — but the pattern underneath it is already on quote sheets landing in Australian inboxes. The label has run miles ahead of the capability.
40%+
Agentic AI projects Gartner expects scrapped
By end 2027 — cost, unclear value, weak controls
$100M
Max penalty for misleading AI claims
Per breach, Australian Consumer Law, from 28 Mar 2026
In Australia, this is a consumer-law problem, not just a hype problem
Overselling an "AI agent" isn't only annoying here — it can be unlawful. The Australian Consumer Law bans misleading or deceptive conduct regardless of the technology behind the claim, the ACCC has named AI-washing an explicit enforcement concern for 2026–27, and maximum penalties for misleading representations doubled to $100 million per breach from 28 March 2026. We've written about the seller's side — how the ACCC made overblown AI claims a $100m problem for the businesses that make them. The buyer's side is the mirror image: if a vendor's "autonomous agent" can't do what the deck promised, you may well have a misrepresentation claim — but a refund two years on doesn't give you back the switching cost or the hours your team lost babysitting it.
Treat it as a cost-intelligence problem. Agent washing is the active cousin of the AI tax quietly creeping onto your software bill: one lifts the price for AI you didn't ask for, the other charges a premium for AI that isn't really there. Both end the same way — money leaving the business for capability it isn't getting. Before you add another "AI" line to the budget, see where AI would actually move the needle in your business, then go shopping for that specific outcome rather than a label.
Already using software that touches this? A 30-minute check maps which of these rules apply to your business.
Book the AI checkFive questions that separate an agent from a rebrand
You don't need to be technical to pressure-test a claim. Ask five things. One: can it both start and finish a task without a person driving each step — or does it just answer when spoken to? Two: does it actually act inside your systems — book the job, send the invoice, update the record — or only chat beside them? Three: will they give you a two-week trial on your own data and workflow, not a polished canned demo? A scripted demo that can't survive your real data is the tell. Four: make them show the failure cases — what it gets wrong, how often, and what a human still has to check. Five: ask exactly what it's allowed to touch and who's accountable when it acts.
A vendor selling a real agent can answer all five plainly, because they've had to build the controls to make it safe. A vendor selling a rebrand will reach for adjectives. The questions cost you nothing and they sort the two apart in a single meeting.
Buy the outcome, not the label
Don't buy "an AI agent". Buy a measured reduction in a specific cost — hours off quoting, calls off voicemail, days off the month-end close — and make the vendor prove it on your data before you sign, not after. The label is marketing; the hours saved are the only number that survives contact with your business. Pick the one workflow that's bleeding time, hold any tool to that single result, and walk if it can't show it.
If you want a second set of eyes on a vendor's claim before you commit, a short call will help you tell the agent from the costume — and point you at the cheaper, duller tools that often return more than the ones with "agent" on the box.
Key takeaways
Common questions
What is "agent washing"?
It's Gartner's term for marketing a product as an autonomous AI agent when it's really a rebranded chatbot, rule-based automation or RPA script that still needs a human to start and steer every step. Gartner estimates only a small fraction of self-described agentic vendors offer genuinely autonomous capability.
What's the difference between an AI chatbot and an AI agent?
A chatbot responds to prompts. An agent acts — it's wired into your systems and takes steps on its own until a task is complete. If a tool can't both start and finish a task without a person driving each step, it isn't really an agent, whatever the label says.
Is it illegal for a vendor to call a chatbot an "AI agent" in Australia?
It can be. The Australian Consumer Law prohibits misleading or deceptive conduct regardless of the technology, and the ACCC has named AI-washing an enforcement concern, with penalties up to $100 million per breach. Whether a specific claim crosses the line depends on the facts — but the exposure is real for the vendor making it.
Sources
Assumptions & methodology
- Gartner's term "agent washing" and the forecast that more than 40% of agentic AI projects will be cancelled by the end of 2027 (citing escalating costs, unclear business value and inadequate risk controls) are from Gartner's press release dated 25 June 2025. The 40% figure is a forecast, not a measured count, and was revisited in a Forbes analysis dated 7 July 2026. Gartner's estimate that only a small fraction of self-described agentic vendors offer genuine agentic capability is reported in that coverage and is presented here as an estimate.
- The Australian Consumer Law's prohibition on misleading or deceptive conduct applying regardless of technology, the ACCC's identification of AI-washing as an enforcement concern for 2026–27, and the increase in maximum penalties for breaches of the relevant provisions to $100 million per contravention from 28 March 2026 are as described in our earlier note on ACCC AI-washing enforcement and in the ACCC's public materials. Whether any particular claim is misleading is a question of fact, untested here.
- Last reviewed 8 October 2026. General information, not legal advice — confirm your obligations, and pressure-test any vendor's claims, before relying on them.
If this applies to you
Not sure the "AI agent" you're being sold is real?
We help Australian businesses separate the AI that returns something from the AI that's just a label — and pressure-test a vendor's claim before you sign. Book a call and we'll look at it with you.