By AIGAS
AIGAS Weekly | 17 September 2026
Research window: 10–16 September 2026; checked Wednesday morning.
AI adoption becomes more useful when we can answer some straightforward questions. Which features are actually available? What information can they access? Who checks the results? And what happens when the technology changes?
This week, Karbon’s product roadmap and two UK public-sector developments give us useful examples to consider. Together, they reinforce a principle behind our AI governance framework for accounting firms: clear controls should help people use AI confidently.
Karbon separates available AI features from what comes next
Karbon’s programme for its 16 September product-roadmap sessions distinguishes between existing period-close features and forthcoming capabilities. It describes AI document validation and a close-check AI rule builder as shipped, while presenting new Kai capabilities as a preview ahead of wider release. This is roadmap information, rather than confirmation that everything shown is available to every firm. Read Karbon’s programme.
For us, that distinction matters. Firms need to make decisions against what they can use today, with the permissions and support available on their own subscription.
Before introducing a feature into client work, we would ask the supplier to confirm access, limitations and the review process. With document validation, for example, what exactly is being checked? What falls outside the check? Who investigates an exception?
A successful demonstration is a useful starting point. Testing the feature against representative work, including awkward cases, helps establish where it belongs in the firm’s workflow.
A government AI gateway offers a practical lesson
On 15 September, the Ministry of Housing, Communities and Local Government published an account of its AI Gateway: a shared departmental platform for accessing approved AI technologies. The department describes governance documentation covering acceptable use, accountability, operational responsibilities and data handling, linked to its AI Register and approval process. Read the MHCLG account.
We see a useful principle here for accounting firms, even where a dedicated technical gateway would be unnecessary.
Make the approved route easy to follow. Staff should know which tools they may use, what information they may enter and where to go when a task falls outside the agreed boundaries.
For a smaller practice, that might mean a short approved-tools list, properly configured business accounts and one named person responsible for new requests. The arrangements should be proportionate, understandable and used in everyday work.
Healthcare proposals highlight why approval needs follow-up
On 10 September, the National Commission into the Regulation of AI in Healthcare published recommendations including staged authorisation and ongoing monitoring of AI-enabled medical devices. The government and MHRA will consider the recommendations. These are healthcare proposals, not new requirements for accounting firms. Read the official announcement.
The transferable lesson, in our view, is that approving an AI tool is the beginning of oversight.
For accounting firms, we would apply that thinking through a defined pilot, a named reviewer and a clear decision about when wider use is appropriate. We would also revisit that decision after significant product changes, a change in the data being processed or an unexpected result.
This gives teams room to learn while keeping responsibility clear. It also creates evidence that the firm is actively managing its use of AI.
This week’s action for your firm
Choose one AI feature already in use and check:
- Availability: Is it released, restricted or still in preview?
- Boundaries: What data and tasks are approved, and who authorises exceptions?
- Review: Who checks its performance, and what would trigger reassessment?
Record the answers in your free AIGAS AI Tool Register and assign an owner to any gaps.
We believe governance should make responsible adoption easier. Knowing what is running, how it is controlled and when it needs another look is a practical place to start.