By AIGAS
AIGAS Weekly | 5 November 2026
Research window: 29 October–4 November 2026; checked Wednesday morning.
AI governance is often discussed as a collection of technical safeguards. This week’s developments show why that is only half the picture.
Accounting firms also need clear ownership, informed challenge and a safe route for people to raise concerns. Our AI governance framework for accounting firms brings those elements together so firms can adopt useful technology with proportionate, demonstrable control.
IRIS puts data location and audit trails into its launch claims
On 2 November, IRIS launched IRIS Document Management in the UK and US. The product uses AI for plain-language document search, optical character recognition and automatic tagging, alongside filing rules and a client portal.
IRIS says client data remains within the firm’s tenant and storage region, is not used to train AI models, and is covered by retention controls, legal holds and audit trails. Read the IRIS announcement. IRIS
These are useful procurement disclosures, but they remain supplier claims. Firms should ask which models and subprocessors perform the AI tasks, what the audit trail records, and how incorrect tags or filing decisions are detected and corrected. The AIGAS Vendor Directory provides a consistent structure for comparing this kind of information.
Dext distinguishes approval from automation
Dext’s UK product update, published on 2 November, describes AI Assist, a coming-soon email scan and auto-publish prompts. Dext says email scan will not send a document to Dext until the user approves it. It also claims that plain-English AI Assist guidance can reduce manual edits by up to 80%. Read Dext’s update. Dext
The approval point is important, but so is the boundary around it. A user approving the transfer of a document is not necessarily approving its classification, accounting treatment or publication.
We would therefore map each decision separately: collection, extraction, categorisation, posting and payment. For every stage, define what is automatic, what is suggested and what requires a person. The 80% figure is a vendor claim; the update does not provide an independent benchmark for it.
Microsoft offers a governance model firms can scale down
On 29 October, Microsoft published an account of its internal AI councils. Its model separates strategy, enablement, data, continuous improvement, responsible AI and measurement responsibilities. Microsoft presents this as the operating structure behind its own AI deployment. Read Microsoft’s case study. microsoft.com
An accounting practice does not need six committees. It does need the functions those committees represent.
In a smaller firm, one monthly AI governance meeting could cover approved use cases, data readiness, incidents, staff feedback and evidence of value. The important point is that commercial enthusiasm, technical delivery and independent challenge should not collapse into one undocumented decision.
DeepMind dispute highlights the importance of speaking up
On 3 November, The Washington Post reported that UK Google DeepMind employees were challenging non-disparagement clauses that they said could restrict public criticism after employment. The newspaper says it reviewed a contract and interviewed two current employees.
Google responded that it does not enforce such clauses after employment, encourages internal dialogue and has removed the post-employment element for new hires. Employees quoted in the report want existing contracts changed as well. Read the report. The Washington Post
This is a reported employment dispute, not a finding that Google acted unlawfully. The wider governance lesson is still relevant: a speak-up route must be trusted in practice. Staff should know how to raise an AI concern, who reviews it and when escalation outside the immediate delivery team is appropriate.
This week’s action for your firm
Review one AI-enabled workflow and check:
- Boundary: Which steps collect, classify, post or share information automatically?
- Evidence: Can supplier claims about accuracy, data use and efficiency be substantiated?
- Challenge: Can staff raise concerns safely, and who has authority to investigate or pause the tool?
Record the tool, controls and owner in your free AIGAS AI Tool Register. Good governance gives firms confidence to move forward because both the technology and the decisions around it can be challenged.