The rule nobody had to write

Search for an AI-specific greenwashing rule and you will not find one. That absence is not a gap. The environmental-claims regimes on both sides of the world were drafted around products and conduct, not technologies, and AI marketing walked straight into them.

This piece sets out what actually applies to environmental claims about AI, with each instrument named precisely, because this is an area where secondary coverage routinely cites the wrong law.

First, the wrong-instrument trap

Two EU instruments share the greenwashing space in popular coverage, and only one of them is law. Directive (EU) 2024/825, the Empowering Consumers for the Green Transition Directive, was adopted on 28 February 2024; Member States had to transpose it by 27 March 2026 and must apply their measures from 27 September 2026. The Green Claims Directive is a different proposal, COM(2023) 166, and it has not been adopted: the European Parliament's procedure file records it as awaiting the Council's first-reading position. Any compliance document that treats "the Green Claims Directive" as binding EU law is citing an instrument that does not yet exist in force.

How AI got inside the regime

The Empowering Consumers Directive works by amending the Unfair Commercial Practices Directive, and the UCPD's definition of product does the work: a product "means any good or service including immovable property, digital service and digital content". An AI assistant, an API, a chatbot subscription: all digital services. Every environmental-claim prohibition in the amended UCPD reaches them exactly as it reaches a bottle of detergent, with no AI-specific provision required.

What is banned from 27 September 2026

Offsetting-based neutrality claims, outright. The new Annex I blacklist entry reads:

"Claiming, based on the offsetting of greenhouse gas emissions, that a product has a neutral, reduced or positive impact on the environment in terms of greenhouse gas emissions."

Blacklist entries are banned in all circumstances, with no case-by-case fairness assessment. So "carbon-neutral AI, powered by offsets" is not a claim that needs better substantiation after 27 September 2026; at product level in the EU, it is a claim that cannot be made at all. The Commission's own FAQ draws the boundary that matters for AI marketing: the prohibition is product-level. A company-level claim resting on offsets is not automatically banned, and is instead assessed under the general misleading-practices rules. An organisation whose "sustainable AI" page mixes product and company claims should know which side of that line every sentence sits on.

Generic claims, unless you can prove excellence. Generic environmental claims are prohibited where the trader cannot demonstrate recognised excellent environmental performance relevant to the claim. The directive's own recital lists examples, and two of them should stop an AI marketer mid-sentence: "energy efficient" and "climate friendly". Those phrases are everywhere in AI product copy, usually with nothing behind them but an intuition that newer chips use less power.

The penalty architecture. For cross-border infringements coordinated under the EU's consumer-protection cooperation machinery, Member States must provide for maximum fines of at least 4% of the trader's annual turnover in the Member States concerned. That is a floor on the maximum each state must make available, not a tariff anyone has been charged; the exposure is real, and it is turnover-linked.

Australia: the enforcement that has already happened

Australia needed no new statute. The ACCC's Making environmental claims: a guide for business sets out eight principles under the ordinary Australian Consumer Law, and the Federal Court has already ordered Clorox Australia to pay A$8.25 million for representing that certain GLAD bags were made from at least 50% recycled ocean plastic. The ACL's maximum penalties per contravention are the greater of A$100 million (the flat limb having doubled on 28 March 2026), three times the benefit obtained, or 30% of adjusted turnover during the breach period.

ASIC has three completed greenwashing court outcomes: Vanguard Investments Australia, A$12.9 million; Mercer Superannuation, A$11.3 million; and Active Super, A$10.5 million. An honest use of those figures requires their limits: all three are financial-services cases about ESG investment screens, where practice diverged from the stated screen, not product-marketing cases. The transferable principle is exact, though: an undisclosed gap between what your sustainability statement says and what your systems actually do attracts eight-figure penalties under ordinary misleading-conduct law. ASIC's own reporting also shows that most of its greenwashing work is quiet corrective disclosure rather than court action, which means the public cases are the floor of the activity, not the whole of it.

The statement no regulator has made, and why it does not matter

We looked for a regulator saying, in terms, that environmental-claims substantiation rules apply to AI. We did not find one. The ACCC's current enforcement priorities treat greenwashing at length and never mention AI. That silence is worth knowing about, and it changes nothing. The closest analogue points the way: announcing its Operation AI Comply sweep against deceptive AI capability claims, the FTC's chair said "there is no AI exemption from the laws on the books". That was said about AI-washing, not greenwashing, but it is the same legal logic the EU reached by defining product to include digital services: general law does not need a technology-specific hook.

The two regimes also compose. A claim like "our carbon-neutral AI assistant" can misdescribe the AI and misdescribe the carbon in a single sentence, and each misdescription has its own enforcement path. For the AI half, see our explainer on AI washing.

What this means for an AI sustainability page

  • Audit for the blacklist before September 2026. Any EU-facing product claim of carbon neutrality that rests on offsets has a hard stop date. Rewrite it or retire it.
  • Treat "energy efficient AI" as a regulated phrase. Generic claims need demonstrable, recognised excellence behind them. An intuition about chip generations is not that.
  • Keep product claims and company claims separate. The offsetting ban is product-level; the general rules govern company-level claims. A page that blurs the two inherits the stricter treatment.
  • Do not launder a boundary change into a reduction. As we set out in the companion piece on cloud emissions, moving AI workloads to a third-party cloud moves the electricity out of your Scope 2 without removing an emission. Presented as decarbonisation, that is the exact shape these regimes exist to catch.
  • Flow the discipline into procurement. If your marketing repeats a vendor's environmental claim, their substantiation problem becomes yours. Ask for the evidence in due diligence, not after the complaint.

This article describes Directive (EU) 2024/825, the amended Unfair Commercial Practices Directive, the Australian Consumer Law and regulator enforcement outcomes as at the dates given. It is general information, not legal advice. Transposition detail varies by Member State, and Australian penalty maxima are per contravention and fact-dependent; take advice on your own claims.

Related reading

Sources: European Commission, Questions and Answers on the Empowering Consumers Directive · Directive (EU) 2024/825 (EUR-Lex) · ACCC, Making environmental claims: a guide for business · ACCC media release, Clorox penalty · FTC, Operation AI Comply announcement