Skip to content
UNTIL 31 DECWhich kind of AI user are you?
Take the two-minute check
AIAC AI ASSURANCE COUNCIL™

AI in finance and accounting

COSO 2013 Framework, Principle 9 — Identifies and analyzes significant change

Principle 9 asks an organisation to identify and analyse changes that could significantly affect its system of internal control. COSO’s 2026 paper reads that onto changes a finance function does not make: a supplier’s model update, a new safety filter, a repointed retrieval source. The control this page describes is contractual — the right to be told before the change lands, not after.

§ 1When it applies

Which editions this rests on

  1. 1 January 2013

    Publication year of COSO’s Internal Control — Integrated Framework, held as the start of the year because the day is not asserted. An edition marker, not a commencement: nothing happens to anyone who departs from a voluntary framework.

  2. 1 January 2026

    Publication year of COSO’s Achieving Effective Internal Control Over Generative AI (GenAI), again held as the start of the year. Commissioned thought leadership rather than an amendment to the Framework; it creates no principle and no duty.

§ 2In practice

The change you did not make

Principle 9 belongs to the Risk Assessment component, and answers a simple observation: a control system designed against last year’s risks stops fitting when something material changes. Nothing about it is binding. COSO’s Framework is voluntary, carries no penalty, and may be departed from without consequence. Its usefulness here is that it already holds the concept a finance function needs and has not yet applied to the tools it buys.

The 2026 paper’s position is that changes to model architecture, training data, prompts, retrieval connectors, safety filters or integration points can materially alter a system’s risk profile; that risk assessment must track them; and that they should trigger re-evaluation before they affect operations rather than after. It expressly reaches supplier-driven changes, and notes that these may call for notification obligations or independent verification. Read that against the factor in R220.8 on the employing organisation’s oversight of the technology across its lifecycle, and the shape of the duty is clear: judging what verification an output needs is impossible if you do not know the tool changed.

For a finance function the operative word in all of that is "before". You cannot re-evaluate a change you learn about from a variance, and almost nothing about a hosted model is within your gift to detect. So the control is not technical, it is contractual: a term requiring the supplier to notify material changes to the model, the prompt scaffolding, the retrieval sources or the safety filters, with enough notice to test before the change reaches production, and a right to the evaluation results behind it. That term is written at procurement, by people usually negotiating price and uptime, and it is the single most consequential thing a finance function can do about AI risk in a close process. Its absence is what turns a supplier’s routine release into an undisclosed change to internal control — the disclosure problem taken up at deploying an AI tool mid-year.

§ 3Who it binds

Which tools this reaches

Accountant in business

Any AI tool a finance function does not itself control — a hosted service, an embedded feature in a finance system, a model consumed through an interface — where the supplier can change behaviour without the customer acting. Principle 9 binds nobody: COSO’s Framework is voluntary and no consequence follows from departing from it. It applies here through the control framework management has already adopted.

§ 4What discharges it

What the change file must hold

The artefacts an assessor asks to see, and what makes each one sufficient rather than merely present.

  1. 01

    The vendor change-notification term, as signed

    The clause itself, not the account manager’s assurance. What counts as a material change, how much notice is owed, and whether the supplier must hand over its own evaluation results with it.

  2. 02

    A log of notifications actually received

    Dates, what the supplier said had changed, and what the function did about each. An empty log after two years of service is evidence about the term rather than about the supplier.

  3. 03

    The re-evaluation performed before a change reached production

    A test on the function’s own population against the prior configuration, dated before the change went live. Afterwards it is an investigation, not a re-evaluation.

  4. 04

    A dated record of which model version produced which output

    Kept so that a variance found in April can be attributed to something that moved in February. Without it, a supplier release and an accounting error are indistinguishable months later.

§ 5Worked example

Worked example — a release nobody mentioned

A finance function uses a hosted AI tool to code and route supplier invoices. The contract is a standard subscription: uptime commitments, data protection terms, a support service level, and a clause reserving the supplier’s right to improve the service. In February the supplier moves to a newer underlying model. The finance team is not told, because nothing in the contract says it should be. Coding accuracy on one category of invoice falls, and the miscoding surfaces in the April management accounts as an unexplained variance.

What went wrong, and at what point was it fixable?

The failure is not the release and not the model. It is that a change to a control the group operates was made by somebody else, and the group had no way of knowing. Principle 9 as the 2026 paper applies it asks for that change to be identified and analysed before it affects operations, and the only mechanism that could have delivered it was a contract term the group did not ask for. Everything downstream follows: two months of miscoded invoices, a variance investigated as an accounting question rather than a technology one, and no way now to say which postings were produced by which model. The fixable point was at signature, and the term is neither exotic nor expensive — notice of material model changes, a stated notice period, and access to the supplier’s own evaluation of the change. The second fixable point is renewal, which usually arrives sooner than a finance team expects.

§ 6What a weak answer looks like

A vendor assessment with a date on it

A vendor assessment performed once, at selection. Security questionnaire returned, references taken, a pilot run on a sample of the function’s own data, sign-off filed. It was good work, and it describes a product that no longer exists in the form assessed. Nothing in the file establishes what the supplier has shipped since, whether the function would have been told, or what the current configuration is — and the assessment’s age is rarely noticed, because a tool that is working generates no reason to look at it.

§ 7Elsewhere

Where the change lands

Where another instrument addresses the same obligation. These are correspondences, not comparisons — the Council does not rank one framework against another.

  • MAS instruments

    MAS instruments reach the same event from a supervised institution’s side, where model change after deployment is treated as a matter of ongoing monitoring rather than of contract terms.

  • Finance and accounting standards

    Whether a supplier’s change matters depends on whether a person re-performs the work behind the control. A non-reliance process absorbs a model change that a reliance process does not.

A correspondence indicates that two instruments address the same underlying obligation. It is not a mapping endorsed by either body, not a statement that one satisfies the other, and not a judgement about which is more demanding.

§ 8Where this is assessed

Where this is assessed

Examined in 2 credentials, in the domains named on each card.

AIAC-FNAFPractice

Certified AI in Finance & Accounting, Fundamentals

Close and reconciliation support · 20% of the paper

For finance and accounting professionals who use AI in analysis, close and reporting work — and verify the output before the numbers are relied on.

Financial servicesEnrolment open
AIAC-FNAPPractice

Certified AI in Finance & Accounting, Professional

Governing AI use across the finance function · 20% of the paper

For senior finance and accounting professionals who set the terms on which AI is used across a finance function — and answer for the estimates, controls and disclosures it touches.

Financial servicesEnrolment open

Certification

Examined on this standard of evidence

What this page sets out is examined, on work the candidate has not seen, in the credentials beside it. Every paper is machine marked against its key and every result is reviewed and released by a named person, and every credential is on the public register.

AIAC-FNAF

Certified AI in Finance & Accounting, Fundamentals

2 hours · remote, proctored · valid 2 years

Enrol

AIAC-FNAP

Certified AI in Finance & Accounting, Professional

2 hours · remote, proctored · valid 2 years

Enrol