Inferred Consent Under the Spam Act: Can AI Agents Use It?
Sometimes, never as a default. Section 16(2) of the Spam Act 2003 (Cth) disapplies the s 16(1) prohibition where the account-holder consented, and Schedule 2 clause 2 permits consent inferred from conduct and business relationships. Section 16(5) puts the evidential burden on the sender. The ACMA expects unsubscribes actioned within 5 business days.
- Inferred consent is a defence, not a permission. Under s 16(5) a person relying on it “bears an evidential burden in relation to that matter”.
- Publication alone is not consent. Schedule 2 clause 4(1) says consent “may not be inferred from the mere fact that the relevant electronic address has been published”.
- Three obligations: consent (s 16), accurate sender identification (s 17), functional unsubscribe (s 18). Business addresses are in scope.
- Penalties are in penalty units: for a body corporate, up to 2,000 units for multiple s 16(1) contraventions on one day, 10,000 units with a prior record (s 25).
The three obligations the Spam Act imposes
Part 2 contains three rules for commercial electronic messages with an Australian link, each a civil penalty provision (ss 16(11), 17(7), 18(8)):
- Section 16 prohibits sending. Subsection 16(2) is the carve-out: it “does not apply if the relevant electronic account-holder consented to the sending of the message”, with a note pointing to Schedule 2. Subsections 16(3) and 16(4) add narrow mistake-based exceptions.
- Section 17 requires the message to identify who authorised it, with contact details “reasonably likely to be valid for at least 30 days after the message is sent” (s 17(1)(d)).
- Section 18 requires an unsubscribe statement “presented in a clear and conspicuous manner” (s 18(1)(d)), using an address that works “at all times during a period of at least 30 days after the message is sent” (s 18(1)(e)).
Verified against Spam Act compilation No. 10 (register ID C2016C00614, compilation date 10 March 2016) on legislation.gov.au, checked 4 September 2026. For SMS, s 17 interacts with sender ID registration.
What inferred consent actually says in Schedule 2
Clause 2 defines consent as express consent, or “consent that can reasonably be inferred from” two things read together: “the conduct” and “the business and other relationships” “of the individual or organisation concerned”. That is the whole definition — no list of qualifying relationships, no safe harbour.
Schedule 2 sets no expiry for inferred consent, and s 16(2) asks whether the account-holder consented to the sending of the message — so the inference has to hold at send time, and a lapsed relationship stops supporting it. Clause 4 then closes the obvious loophole. Clause 4(1): consent “may not be inferred from the mere fact that the relevant electronic address has been published”. Clause 4(2) makes one narrow exception, the conspicuously published work-related address, needing four things at once: the address enables the public to reach a particular employee, director, officer, partner, office-holder, self-employed individual, or an individual or group performing a particular function or role; it “has been conspicuously published”; it is reasonable to assume publication occurred with that person’s or organisation’s agreement; and the publication carries no statement declining unsolicited commercial messages. Even then, the deemed consent covers only messages “relevant to” that person’s work-related business, functions or duties. A scraped info@ address is a poor fit for the first limb — it addresses the organisation at large rather than a particular person, office, function or role — and a generic pitch fails the last.
Clause 5 lets regulations either restrict (cl 5(1)) or expand (cl 5(2)) when consent may be inferred. The Spam Regulations 2021 (F2021L00285, as made, in force from 1 April 2021, no compilations) contain no such regulation: their only substantive provisions are s 6 (faxes, under s 6(7) of the Act) and s 7 (unsubscribe-address conditions, under s 18(1)(g)). Full instrument read, checked 4 September 2026.
The sender carries the burden, and that is the whole problem
Section 16(5) is one sentence: “A person who wishes to rely on subsection (2), (3) or (4) bears an evidential burden in relation to that matter.” The ACMA puts it plainly on its Avoid sending spam page: “Under the Spam Act, it’s up to you to prove that you have a person’s consent.” That page also states “Inferred consent is not as reliable as getting someone’s express consent” (checked 4 September 2026).
The ACMA’s 2024 Statement of Expectations on the use of consent names the artefact: “Records should include the method by which the consent was obtained, the terms that applied and the date and time it was obtained.” Where the ACMA compels those records, “they must be produced”. It also rejects one inference outright: do not infer consent “from one-off purchases by consumers, including in cases where phone numbers or email addresses are obtained to send receipts or tax invoices”. Quotations taken from the DOCX linked on that page, checked 4 September 2026.
Now add autonomy. A human sending 40 emails a day makes 40 decisions someone could reconstruct. An agent working across more than 10,000 leads a day and making 28x more contact attempts makes consent decisions nobody reviews before the send. That does not change the law. It changes whether you can answer the ACMA six months later.
Consent basis, evidence, and what an agent can establish unaided
| Consent basis | Evidence you must produce | Can an agent establish it unaided? |
|---|---|---|
| Express consent (form, tick box, phone, in person) | Who consented, the terms shown, the channel, the date and time | No. It consumes a record; it cannot create one, and s 16 stops it messaging to ask. |
| Inferred from an ongoing relationship (Sch 2 cl 2) | That the relationship was current at send time, plus a link between the product and that relationship | Only if the CRM exposes relationship state and product scope as fields checked at send time. |
| Inferred from a one-off purchase or receipt address | Nothing that works; the Statement of Expectations lists it as a practice not to use | No. Block it in configuration, not the prompt. |
| Conspicuously published work address (Sch 2 cl 4(2)) | The publication and where it sits, that the address maps to a named person or role, that publication was plausibly with their agreement, that it carried no anti-spam notice, and that the message suits that role | Partly. It captures the source URL and a snapshot; a human signs off relevance. |
| Purchased or third-party list | The underlying consent for each address | No. It inherits whatever the vendor can evidence, which is usually less than the contract implies. |
The record an agent should write at send time
The fix is not a better prompt. It is a record the sending system writes in the same transaction as the message:
- Consent basis — express, inferred-relationship, or Sch 2 cl 4(2). Null blocks the send.
- Source — form ID, CRM record, contract, or the exact URL the address was published on, plus a snapshot.
- Timestamp — when consent was obtained, and separately when the message went out.
- Exact wording shown — the verbatim consent text, not a pointer to a template since edited.
- Relationship relied on — which account or engagement, and why the product sits in its scope.
- Unsubscribe state — suppression checked at send time, plus the s 18 address and its expiry, and the s 17 identity block as rendered.
Zian’s agents check consent state at send time rather than list-build time, and a missing basis suppresses the contact rather than degrading to a guess. Ask any vendor to show you one such record for a real send. Apply For Partnership to work through it.
Penalties: read them in penalty units, not stale dollars
Section 25 sets maximums in penalty units. A body corporate with no prior record faces up to 100 units per s 16(1) contravention and 2,000 units for two or more contraventions on one day; with a prior record, 500 and 10,000. Breaches of ss 17 and 18 are capped at half those. The Federal Court sets the actual figure under s 24, having regard to the matters in s 24(2).
Dollar conversions of those caps circulate widely, and most were worked out against an older penalty unit value. We are not adding another, because the multiplier is genuinely unsettled. The authorised text of s 4AA(1) of the Crimes Act 1914 still reads A$330 (compilation No. 167, compilation date 27 August 2026). Indexation under s 4AA(3) fell due on 1 July 2026, and the Crimes (Amount of a Penalty Unit) Instrument 2026, made under s 4AA(1A), publishes the indexed amount as A$364 from that date. The instrument’s own note then limits it: the indexed amount “only applies to offences committed on or after 1 July 2026 (see subsection 4AA(8) of the Act)”. Spam Act contraventions are not offences — s 27 provides that “Criminal proceedings do not lie against a person only because the person has contravened a civil penalty provision”. Whether a 10,000-unit cap converts at A$330 or A$364 is a question for a lawyer, and across that cap the two answers are A$340,000 apart. Cite the units, not the dollars. All three instruments read on legislation.gov.au, checked 4 September 2026.
Where the Spam Act stops and the Do Not Call Register Act starts
The Spam Act covers commercial electronic messages — email, SMS, instant messaging — and s 5(5) carves out any message “sent by way of a voice call made using a standard telephone service”. Voice sits under the Do Not Call Register Act 2006 (Cth), and the ACMA’s Statement of Expectations is explicit that “Commercial calls made using an auto-dialler (robocalls) or that use pre-recorded or AI voices are all telemarketing calls.”
Two differences matter for a multi-channel agent. The Register excludes numbers primarily for business use, whereas the Spam Act applies to business email and SMS; and the two consent regimes are separate instruments, so satisfying one does not satisfy the other. We cover the voice side in the Do Not Call Register and AI voice agents, the cross-border picture in outreach compliance for AI agents, and records in rewriting your consent language.
Frequently asked questions
Can an AI agent rely on inferred consent under the Spam Act?
It can, on the same terms as a human sender: the Act regulates the message and the person who authorised it, not the tool. But s 16(5) puts the evidential burden on the sender, so the real question is whether your agent writes a record at send time that would satisfy the ACMA later. If it does not, do not rely on inference.
Is a published email address on a company website consent?
No. Schedule 2 clause 4(1) of the Spam Act says consent “may not be inferred from the mere fact that the relevant electronic address has been published”. Clause 4(2) creates a narrow exception for a conspicuously published work-related address, but only where it maps to a particular person or role, publication was plausibly with their agreement, it carries no anti-spam notice, and your message is relevant to that role’s duties.
How quickly must an unsubscribe be honoured?
The ACMA’s guidance says the unsubscribe option must honour “a request to unsubscribe within 5 working days” and be “functional for at least 30 days after you sent the message”. Schedule 2 clause 6 of the Act sets the same period: withdrawal of consent takes effect at the end of 5 business days beginning on the day the unsubscribe message was sent.
Does the Spam Act apply to B2B email in Australia?
Yes. The ACMA’s Statement of Expectations says that for e-marketing “consent must be obtained before messages can be sent, including to businesses” — unlike the Do Not Call Register, which excludes numbers primarily for business use.
Who is liable if an outsourced platform sends the message?
You are, alongside the sender. Section 16(9) extends liability to aiding, procuring or being knowingly concerned in a contravention, and the ACMA says businesses stay responsible “regardless of whether they outsource” their e-marketing or consent gathering.
Build the record before you build the volume
Inferred consent survives an ACMA enquiry only when someone can put the artefact on the table. An agent does not create it — it inherits one, or manufactures exposure at speed. Decide the basis per contact, store it per send, make a missing basis a hard stop. Apply For Partnership.
General information, not legal advice.