In short
Management selects AI vendors and the board sets guardrails
- Management selects AI vendors. The board sets the rules that stop management from choosing the wrong ones.
- Ask what each vendor changes in the work. A vendor should help remove, automate or redeploy work in a measurable way, or the evaluation is incomplete.
- Every AI vendor contract creates a dependency. Boards should require exit strategies and data portability terms before approving strategic vendor relationships.
- For organisations the Privacy Act covers, vendor data handling is a board-level concern. If an AI vendor receives personal information offshore, Australian Privacy Principle (APP) 8 on cross-border disclosure may apply. If nobody has assessed that, the board has a governance gap.
Detail
AI vendor thresholds that bring decisions to the board
Selecting AI vendors is management's job. The board sets the framework that governs how vendors are evaluated, contracted and monitored.
Start with what the vendor changes in the work. When management recommends a vendor, the board should ask whether it removes unnecessary work, automates a manual process or frees people for higher-value tasks. The business case is ready only when the pitch links capability and potential clearly to the work.
Thresholds should trigger board involvement, so the board reviews only the purchases that meet them. Common triggers are an annual contract value above a level the board sets, multi-year commitments and vendors with access to customer or employee data. Others are vendors that provide business-critical capability and any vendor that creates a single point of failure. Below those thresholds, management decides within approved governance frameworks.
For vendors above the threshold, boards should require 5 things from management. The first is an analysis of alternatives: several genuinely evaluated options, scored against defined criteria, with sole-source cases only in exceptional circumstances. The second is a data sovereignty assessment showing where data goes, who can access it and how this meets the Privacy Act. The third is a lock-in analysis of today's switching cost and how it grows over the contract. It should also confirm that data can be exported in usable formats. The fourth is an exit strategy for vendor failure, a large price rise or an acquisition. The fifth is contractual protection: liability terms, audit rights, deletion of data at termination and a ban on using your data for model training.
The Australian context adds specific requirements. For organisations the Privacy Act covers, its obligations stay with the organisation when a vendor handles personal information for it. If you disclose personal information to an overseas vendor, section 16C can make your organisation accountable for how the vendor handles it. Entities regulated by the Australian Prudential Regulation Authority (APRA) have additional requirements for managing service provider risk. ASX continuous disclosure obligations may be triggered by a material AI vendor dependency or failure.
Boards should keep a strategic AI vendor register. It lists every vendor above the threshold, with its risk rating, contract terms and dependency assessment. The board reviews it alongside the Board AI Risk Register each quarter.
Commercial impact
AI vendor dependencies can last for years
AI vendor decisions create dependencies that can last for years. Switching costs in AI are higher than for traditional software, because your data, workflows and staff skills all become tied to one platform. A poor vendor choice wastes money, and it also constrains your AI strategy for the life of the contract.
Data sovereignty is a risk boards can underestimate. Many AI vendors process data on overseas infrastructure. Some use customer data to train their models, and some keep data after the contract ends. A board that has not reviewed the data terms in its AI vendor contracts may not know whether it meets its Privacy Act obligations.
Vendor concentration creates strategic vulnerability. Consider an organisation whose AI capability runs largely through one vendor. When that vendor raises prices at renewal, the organisation has no negotiating position. The switching cost is large, and so is the disruption. A board-level review can find that concentration well before renewal.
The competitive effects are significant. Vendor choices shape which capabilities you can deploy, how quickly you can innovate and whether you can differentiate. A board that treats AI vendor selection as routine procurement misses that strategic dimension.
Regulation adds urgency. The Office of the Australian Information Commissioner (OAIC) has published its regulatory action priorities for 2025-26. They name practices that erode privacy rights in the application of artificial intelligence.
From 10 December 2026, APP 1.7 and APP 1.8 set privacy policy disclosure rules for some automated decisions. Our article on what APP 1.7 requires sets out the test in full. Your vendor must be able to say how its system is involved in decisions about people and what personal information it uses. You need that answer to write the privacy policy description, and a missing answer becomes a board concern.
In practice
Illustrative cases of board review of AI vendor contracts
Illustrative composite: a board asks for an analysis of alternatives for a proposed annual contract for an AI analytics platform. Management first recommends the most expensive option, from a global vendor. The analysis finds a local provider with equivalent capability, Australian data hosting and a lower cost. It also commits in the contract never to use the data for model training.
Illustrative composite: an audit committee reviews an AI vendor contract. It finds a clause that lets the vendor use aggregated customer data to improve its models. The board has the clause removed and adds annual audit rights, so it can check how its data is handled after any breach.
Illustrative composite: a mid-market insurer signed a multi-year AI contract without board review, because the contract fell below the threshold. By the second year, claims processing depended on the vendor. When the vendor doubled its price, the insurer had no exit strategy and no data portability. The board then lowered the review threshold for AI vendors.
Illustrative composite: a healthcare provider's board requires a data sovereignty assessment for every AI vendor processing patient information. The preferred vendor routes data through infrastructure in Singapore. The board judges that an unnecessary risk because Australian-hosted alternatives exist, and accepts a higher price for domestic hosting.
Illustrative composite: another board asks management to show, for each vendor proposal, which work the vendor changes and the projected effect in dollars. Proposals go back to management until they answer clearly.
What to do
Steps to govern AI vendor selection and dependency risk
- Define clear thresholds that trigger board involvement in AI vendor decisions, based on contract value, data access, business criticality and dependency risk.
- For every vendor above the threshold, require an analysis of alternatives, a data sovereignty assessment, a lock-in analysis, an exit strategy and contract protections.
- Ask what work each vendor removes, automates or redeploys. If the answer is unclear, the business case needs more work.
- Keep a strategic AI vendor register with risk ratings and dependency assessments, and review it each quarter with the Board AI Risk Register.
- Set lower review thresholds for AI vendors than for other software, because their switching costs and data risks are higher.
Last reviewed
Put this topic against your own numbers, owners and timelines.