Organisations buying consequential AI services need an accountable human lead to coordinate the negotiation. Procurement, legal, security and technical teams may otherwise agree terms that solve separate problems but do not add up to a workable service.
Key points
- Give the lead a mandate, spending limits and access to specialists.
- Compare total operating and migration costs on the same workload.
- Record declined supplier requests and residual decisions.
A coordination role with defined limits
The AI diplomat in this briefing is a person coordinating technology negotiations. The role does not replace counsel, engineering or executive approval. It connects their requirements and establishes who can make each commitment.
Use seven negotiation headings: data permissions; retention; portability; model changes; cost and capacity; jurisdictions; incidents and liability. Under each, record the request, supplier response, acceptable fallback, evidence and decision owner.
Worked example: price the right to leave
Assume two million completed requests a month. Supplier A charges €0.010 per request; B charges €0.012. Monthly service costs are €20,000 and €24,000, an annual difference of €48,000.
Assume exit requires €90,000 in rebuilding for A and €30,000 for B. If migration occurs within one year, A’s additional €60,000 rebuilding cost outweighs its €48,000 service saving by €12,000. If no migration occurs, A retains the price advantage.
These are hypothetical inputs. They assume equivalent approved quality, capacity and data handling. They are not supplier quotations or a probability estimate. The negotiation can now ask which term or design change reduces the €60,000 gap.
Three distinctions change the terms
OpenAI’s documentation distinguishes default API monitoring retention of up to 30 days from endpoint-specific application state and exceptions. A restriction on training use does not settle those questions. [1]
The European Commission says Data Act switching charges, including relevant data-egress charges, are removed from 12 January 2027. Coverage and exceptions need service-specific legal review. The rule does not remove the customer’s engineering and testing costs. [2]
The EDPB says qualifying transfers outside the EEA must satisfy GDPR Chapter V in addition to ordinary GDPR obligations. A favourable commercial term cannot resolve an unlawful processing arrangement. [3]
Close the operational questions
Before approval, test export, deletion, peak demand and an alternative service. Identify who can suspend access after an incident or material change. Record supplier requests that were refused.
Report cost per accepted output, migration time and cost, unused commitments and unresolved exceptions. The objective is an operating arrangement the institution can support, rather than a discount assessed in isolation.
Sources & research notes
- OpenAI / Data controls (accessed 11 November 2026)Provider-policy snapshot. Product, endpoint, account settings and contractual exceptions matter.
- European Commission / Data Act explained (accessed 11 November 2026)Chapter VI covers switching between data-processing services; applicability and exceptions require service-specific review.
- EDPB / International data transfers (accessed 11 November 2026)Chapter V conditions apply in addition to ordinary GDPR obligations for qualifying transfers.
Independent TACITUS analysis. Research checked 11 November 2026. Published findings retain their dates and limits. Proposed curricula, tests, schedules and worked calculations are labelled analytical examples, not client results.
Continue reading
Training diplomats for AI, and AI for diplomacy