ASIC market integrity rule amendments, made on 15 September, registered on 17 September and released with consultation CS 63 on 24 September 2026, would require governance, testing and seven-year records for trading algorithms, AI-driven or not, from 18 March 2028, ASIC says

What Rule 5.6.3B would require once it commences

Item 69 of Schedule 1 to Instrument 2026/574 inserts Rule 5.6.3B into the Securities Markets rules. The instrument is registered but does not commence until 18 months after registration, so nothing in the new rule applies yet. Once it takes effect, a Trading Participant would need appropriate controls and governance arrangements, including written procedures, for the development, testing, approval, deployment and monitoring of all trading algorithms it uses or makes available to a client.

Once in force, subrule (2) would add a testing trigger at three points: before the participant uses an algorithm for the first time, before it first makes one available to any client, and before it makes a material change. The test is framed as taking reasonable steps to check that the algorithm will not interfere with the efficiency and integrity of a market, a crossing system the participant operates, or the proper functioning of a trading platform.

Subrule (3), also not yet in force, uses a softer verb: development and testing "should be conducted by persons who are suitably qualified to do so". Subrule (4) would require records demonstrating compliance with subrules (1) and (2) to be kept for seven years. Instrument 2026/575 inserts the same wording into the Futures Markets rules as Rule 2.2B.4, on the same 18 month commencement.

Where AI enters the reform, and where it does not

We searched the operative text of both instruments for artificial intelligence and machine learning and found neither term. The new definition, which like the rest of the amendments does not commence until 2028, is functional. A Trading Algorithm is "a computer algorithm which automatically determines with limited or no human intervention, one or more parameters of an Order", such as whether to initiate it, its timing, price or quantity, or how to manage it after submission.

Once the definition commences, four uses would be carved out: systems or processes used only for routing orders to trading platforms, for submitting orders with no determination of order parameters, for producing confirmations, or for post-trade processing.

The AI rationale is in the explanatory statement. It lists addressing "emerging risks, including those from artificial intelligence (AI) and machine learning (ML)" as one of three aims, and devotes three paragraphs to AI risks including herding, flash crashes, opacity and biased data. ASIC's release describes the reforms as "technology neutral" and says they address risks from all trading algorithms, "including those enabled by AI and machine learning". Our view, which is ours and not ASIC's wording, is that an AI model setting order parameters would be caught by function once the rules commence.

The agentic AI change to the false or misleading rule

Paragraph 29 of the explanatory statement says ASIC consulted industry bodies AFMA, FIA and SIAA on an additional change, now part of an instrument that has not commenced, "to address the potential use of agentic AI trading". It says the 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 replacement Rule 5.7.1 is also subject to the 18 month delay. Once it commences it would split principal trading, in subrule (1), from trading on account of another person, in subrule (2), and each would carry an effect limb alongside the intention limb. For client orders, the effect limb in subparagraph 5.7.1(2)(c)(ii) turns on whether the participant ought reasonably suspect the order will have the effect of creating a false or misleading appearance, taking into account the matters in Rule 5.7.2.

The rule text does not mention agentic AI. The link between the effect limb and agentic systems is ASIC's stated intention in the explanatory statement, about a provision not yet in force, and how ASIC would later apply it is not something the instrument settles.

Kill switches, monitoring and the rest of Part 5.6

The amendments also rewrite Part 5.6 as Trading System Requirements, again from commencement and not before. Replacement Rule 5.6.3(1)(c)(ii) would require controls enabling immediate suspension, limitation or prohibition of the operation of one or more trading algorithms. The explanatory statement calls this extending "kill switch" controls to trading algorithms, so a malfunctioning algorithm can be stopped without suspending an entire trading system.

From commencement, new Rule 5.6.3A would require monitoring systems able to immediately identify messages likely to interfere with market efficiency or integrity, post-trade surveillance for false or misleading appearance, and seven-year records of what was identified and the measures taken.

On ASIC's own account, parts of this codify existing expectations. The explanatory statement says there are currently "no specific obligations for securities and futures market participants relating to algorithmic trading", only limited guidance in Regulatory Guide 241, and that the amendments, once in force, would formalise long-standing expectations set out in that guidance. It also says the Office of Impact Analysis advised the likely impact would be no more than minor.

The 2028 date, and the consultation running now

Section 2 of each instrument says it "commences on the day that is 18 months after the day it is registered on the Federal Register of Legislation." Both were registered on 17 September 2026. ASIC's CS 63 page states that "The rules commence on 18 March 2028 following an 18-month transition period." We attribute that date to ASIC and did not compute it ourselves. The Federal Register's page for F2026L01223 shows an "In force" label; we read that as describing the registered instrument, since section 2 defers commencement, and did not investigate what the label denotes.

ASIC's release says it extended the transition to 18 months after considering feedback. It gives no figure for the original proposal in the text we read.

CS 63 is a separate, non-binding step. It seeks comment on draft updates to RG 265 and RG 266, which would absorb relevant RG 241 guidance so that RG 241 can be retired, and it closes at 5pm AEDT on 5 November 2026. ASIC said the proposed changes would reduce the volume of relevant guidance for securities participants by almost 60%; we did not measure that. CS 63 describes the retirement of RG 241 as a proposal.

What we did not verify

What we opened: the as-made PDF text of Instrument 2026/574 (F2026L01223) and its explanatory statement, and the as-made PDF text of Instrument 2026/575 (F2026L01224), all downloaded from the Federal Register of Legislation; ASIC media release 26-226MR; and the CS 63 consultation page. The Register's HTML text page loads its content by script, so we read the downloadable PDFs, and we checked the pull quote character for character against the Register's own text document for F2026L01223. The media release body carries no date line; we take the 24 September 2026 release date from the CS 63 page, which states it.

What we did not open: the explanatory statement for the futures instrument, the draft RG 265 and RG 266 and their summaries, the marked-up rule compilations attached to CS 63, Consultation Paper 386 and the submissions to it, RG 241, and the current unamended text of Rule 5.7.1, so we do not say which parts of the new false or misleading rule are substantively new.

What we refuse to claim: that any of these rules is in force now; that the rules name or single out AI; that the agentic AI intention in the explanatory statement binds how ASIC or a court would read Rule 5.7.1; that the rules impose duties directly on third-party algorithm developers, since the text addresses trading and market participants; and any commencement date other than the one ASIC states.

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.

Related briefings

• IOSCO issues its final AI supervisory toolkit for capital markets
• APRA and ASIC on AI threat sharing and the ACCC authorisation route
• FINRA 2026 on AI agents and deepfake ID risk
• Asia-Pacific AI regulation tracker
• AI Regulation News hub