ASIC added an effect-based limb for agentic AI to its trading rules: from March 2028, market participants must consider a client order’s effect, not just intent

If you run a broking or trading business, or you build software that lets an AI agent place orders through one, ASIC has changed the question your broker must ask of each client order. Under the amended ASIC Market Integrity Rules (Securities Markets), the question is no longer only whether the client may have meant to mislead the market. It is also whether the order will do it. The rules bind market participants, which ASIC’s explanatory statement describes as ranging from large investment banks to small stockbrokers and principal traders: 58 in the securities markets and 38 in the futures markets, 14 of them in both.

ASIC’s reason is in its summary of feedback: Given recent developments in AI, including agentic AI systems that may initiate, adapt or optimise trading activity in ways that are not directly specified by a person at the time an order is generated, ASIC has amended the rule to include an effect-based limb.

What the rule says from March 2028

The false or misleading appearance rule, Rule 5.7.1, already had an effect test when a participant trades as principal, on its own account. For orders it places for someone else, the rule in force today turns on intention: the participant must not act where it ought reasonably suspect that the person has placed the Order with the intention of creating a false or misleading appearance. The amendment keeps that limb and adds a second one beside it.

Intention limb, Rule 5.7.1(2)(c)(i)

The participant ought reasonably suspect that: the person (or if the person is acting on behalf of another person, that other person) has placed the Order with the intention of creating
Status: In the rule today, redrafted

Effect limb, Rule 5.7.1(2)(c)(ii)

The participant ought reasonably suspect that: the Order placed will have the effect of creating
Status: New, from 18 March 2028

Both limbs end the same way: a false or misleading appearance of active trading in any financial product, or with respect to the market for, or the price of, any financial product. Either way the participant must take into account the matters listed in Rule 5.7.2, which now ends with any other relevant matter. The futures rules get the same two limbs in Rule 3.1.2. Source: ASIC Market Integrity Rules (Securities Markets) Amendment Instrument 2026/574, item 93 (PDF page 21); the current compilation for today’s wording.

ASIC’s explanatory statement says what the new limb is for: Our additional amendment is intended to apply where an order generated by an agentic AI system has the effect of creating a false or misleading appearance, whether or not a person intended that effect. The feedback summary adds a qualifier participants will want to read twice: This amendment is not intended to introduce new obligations beyond what a market participant should already be doing to prevent trades that create false or misleading appearances.

Not in the consultation paper

ASIC consulted publicly on these rules in Consultation Paper 386, released on 27 August 2025, with comments closing on 22 October 2025. Its proposal for client orders, B7, was the intention test: an order must not be placed if the participant ought to reasonably suspect that the person has placed it with the intention of creating a false or misleading appearance. The paper does not use the word agentic. The explanatory statement records what happened next. After the formal consultation, ASIC held further bilateral talks with respondents. Then: Given the recent rapid developments in AI, we also consulted with AFMA, FIA and SIAA on an additional change to the false or misleading rule to address the potential use of agentic AI trading. The futures statement and the feedback summary describe the same round as a number of stakeholders, including those three industry bodies.

Proposed in August 2025, made in September 2026
Item	CP 386 (Aug 2025)	Made (Sep 2026)
Client orders: what the participant must suspect	Intention only B7	Intention or effect 5.7.1(2)(c)
Transition before the rules commence	12 months D1	18 months, to 18 March 2028
Monitoring of trading messages	Identify “in real time” B5	“Immediately identify”; real-time removed 5.6.3A
Independent validation of testing	Asked as a question B3Q1	Not required
Testing a trading algorithm before first use and material change	Proposed	Required 5.6.3B

Proposal numbers are CP 386’s (B7 at PDF page 28, D1 at page 36, B5 and B3Q1 in section B); rule numbers are the amended Securities Rules. “Not required” is from ASIC’s summary of feedback (PDF page 4): we have decided not to require independent third-party testing. Sources: CP 386; summary of feedback to CP 386; Amendment Instrument 2026/574 and its explanatory statement.

ASIC received nine non-confidential and seven confidential submissions. Its summary says Most respondents supported our proposed reforms but wanted more guidance, and most of the softening in the table came from that feedback: a longer runway, no “real-time” standard, no mandatory third-party testing. The effect limb went the other way. Of the five changes the explanatory statement lists as made in response to feedback, it is the only one that adds a test; the others ease or clarify a rule. And the published submissions were written to a paper that did not contain it.

Who tests the agent

The new testing rule, 5.6.3B, requires a participant to take reasonable steps to test a trading algorithm before it uses it for the first time, first makes it available to a client, or makes a material change to it. A trading algorithm, in the rules’ new definition, automatically determines one or more parameters of an order with limited or no human intervention. An AI agent that decides whether, when and how much to trade fits that description. What the participant must test depends on whose algorithm it is, and ASIC’s draft updated Regulatory Guide 265, still out for comment, spells out the split.

The participant’s algorithm, or one it provides to a client

Test before use: Required, before first use, before first making it available to a client and before a material change (Rule 5.6.3B(2))
Orders it generates: Both limbs of Rule 5.7.1 apply

A client’s own algorithm or agent

Test before use: The requirement does not apply (draft RG 265.110). The participant should consider testing the trading algorithmic flow against the trading participant’s trading system filters (draft RG 265.111)
Orders it generates: Both limbs of Rule 5.7.1 apply

The dashed card is the case the participant does not have to test. RG 265.110 and 265.111 are draft guidance open for comment until 5 November 2026, not rules. Sources: Amendment Instrument 2026/574, Rule 5.6.3B (PDF page 15); draft updated RG 265, Attachment 1 to CS 63 (PDF page 31).

The draft guide gives one example of the effect limb at work, and it is the agentic case. After saying a participant must not give effect to an order it ought reasonably suspect will create a false or misleading appearance of active trading, whether or not a person intended that, it adds: For example, where the order has been generated by an agentic AI system which has initiated, adapted or optimised trading activity outside of the person’s initial specifications. The same draft says a participant’s monitoring and controls for the two limbs should be the same for both the ‘effect’ and ‘intention’ limbs of the rule.

Why ASIC says it is acting now

The explanatory statement estimates that algorithmic trading is about 85% of all trading in Australian listed equities markets, 94% of trading in SPI 200 futures and 46% in 3-year Treasury bond futures. It gives no period or method for those estimates. Commissioner Simone Constant, in ASIC’s 24 September release, said trading algorithms can also behave in ways that are opaque, unpredictable and potentially manipulative.

In some circumstances, a person may not have intended to create a false or misleading appearance, even where the trading activity has that effect.

ASIC, Explanatory Statement to Amendment Instrument 2026/574, paragraph 29 (PDF page 6)

The release calls the reforms technology neutral, and the rules bear that out: the effect limb never mentions AI. It applies to any client order, however it was generated. Agentic AI is the reason ASIC gives for the limb, in the explanatory statement, the feedback summary and the draft guide, not a condition written into the rule.

The dates

27 Aug 2025CP 386 released, proposing an intention test for client orders and a 12-month transition
22 Oct 2025CP 386 comments close
9 Sep 2026Date of the Minister’s written consent to the securities amendments, which the Corporations Act requires
17 Sep 2026Both amendment instruments registered on the Federal Register of Legislation
24 Sep 2026ASIC release, summary of feedback and CS 63 draft guidance published
5 Nov 2026CS 63 comments close, 5 pm AEDT
18 Mar 2028The amended rules commence

Dates as the sources give them: CP 386 dates from its ASIC page; the consent from the explanatory statement (PDF page 16); registration from the Federal Register; 24 September, 5 November and 18 March 2028 from the CS 63 page. The instruments commence 18 months after the day they were registered.

Our view

The effect limb is the right idea. When an agent adapts its own strategy, asking whether its owner meant to mislead the market can have no useful answer, and a test that looks at what the order does is the honest response to that. ASIC also kept it technology neutral, so the rule does not have to decide what counts as agentic.

But it is the one change made after consultation that adds a test, and it was never put to the public round. ASIC says it is not intended to introduce new obligations beyond what participants should already be doing; if so, it was a clarification that could have gone in the paper, and if not, the public never saw it before it was made. The fix is cheap and still available. CS 63 is open until 5 November and contains the agentic example in RG 265.133. Anyone who will have to apply that example, or who builds the agents it describes, should send ASIC’s markets consultation inbox a view on it now, while the wording can still change. The rule itself does not commence for more than 17 months.

How we did this

On 9 October 2026 we read ASIC’s 24 September release, the CP 386 and CS 63 pages, the 8-page summary of feedback, the relevant parts of CP 386 (proposals B3, B5, B7 and D1), both amendment instruments and their explanatory statements from the Federal Register, Rule 5.7.1 in the current compilation of the Securities Rules (compilation date 15 October 2024), and the AI, trading algorithm and false or misleading appearance sections of the 198-page draft RG 265. We full-text searched CP 386 for “agentic” and found no match. The only arithmetic is ours: 58 plus 38, less the 14 in both, is 82 distinct participants, which we have not written as a total because ASIC does not give one. The instruments say they commence 18 months after registration on 17 September 2026; we use ASIC’s own date, 18 March 2028. We did not read the nine published submissions in full, because they respond to a paper that did not contain the effect limb.

“Our view” is opinion based on the documents cited. We have not asked ASIC about anything here.

Sources

ASIC, 26-226MR ASIC strengthens AI trading safeguards and streamlines market integrity rules, media release, 24 September 2026 (read 9 October 2026): technology neutral; the Commissioner’s statement; the 5 November submission date.
ASIC Market Integrity Rules (Securities Markets) Amendment Instrument 2026/574, F2026L01223, registered 17 September 2026 (PDF, read 9 October 2026): Rules 5.6.3A, 5.6.3B and 5.7.1; the trading algorithm definition; commencement.
ASIC, Explanatory Statement, ASIC Market Integrity Rules (Securities Markets) Amendment Instrument 2026/574, 19 pages (read in full 9 October 2026): the agentic AI consultation (paragraph 29), the 58 participants, the algorithmic trading estimates, the Minister’s consent.
ASIC Market Integrity Rules (Futures Markets) Amendment Instrument 2026/575, F2026L01224, with its explanatory statement (read 9 October 2026): Rule 3.1.2; the 38 futures participants, 14 also in the securities market; the futures account of the agentic AI consultation.
ASIC Market Integrity Rules (Securities Markets) 2017, current compilation (compilation date 15 October 2024, read 9 October 2026): Rule 5.7.1 as it stands today.
ASIC, CP 386 Proposed amendments to the ASIC market integrity rules: Trading systems and automated trading, 27 August 2025, with the 47-page paper (read 9 October 2026): proposals B3Q1, B5, B7 and D1; the release and closing dates; the submission counts.
ASIC, Summary of feedback to CP 386 and ASIC’s response, September 2026, 8 pages (read in full 9 October 2026): the effect-based limb and its reason; no third-party testing; the 18-month transition.
ASIC, CS 63 Proposed updates to RG 265 and RG 266 and withdrawal of RG 241, 24 September 2026, with Attachment 1, draft updated RG 265 (read 9 October 2026): the 18 March 2028 commencement; the 5 November closing date; draft RG 265.110, 265.111, 265.131 and 265.133.

Read the rules differently, or trade through an AI agent and have a view on the effect limb? Tell us and we will check it against the documents and log the outcome here.