Direct answer
For every external component, record the provider, product, version, purpose, data supplied, output received, permissions, dependencies and accountable owner. The organisation keeps visible its own responsibility for selecting and configuring the service, transferring data, using an output in a decision or publication, and monitoring problems. Record the provider's responsibility separately within the contract, technical documentation, service scope and applicable law. 'The provider did it' is not an accountability matrix.
In plain language
Imagine using a rented kitchen. The oven's owner may be responsible for faulty equipment, but you remain responsible for the ingredients you choose, how you prepare the food and what you serve to the customer. An external tool does not move every decision outside your organisation.
Why this matters
A change by one provider in the chain may alter data flows, model behaviour, security or the meaning of an output. Without an inventory and exit plan, the organisation may be unable to identify the responsible component or continue the service safely.
Do not confuse
- The provider supplies the product; the deploying or using organisation applies it to a particular purpose.
- The technical operator, decision owner and legally accountable party are not always the same.
- A dataset, foundation model, fine-tune, API, plug-in and infrastructure service are separate supply items.
- A contractual allocation of responsibility does not automatically remove every obligation owed to affected people.
- The availability of a service does not prove that it is safe or suitable for a particular use.
What should you do?
- Inventory every model, dataset, API, plug-in, tool and infrastructure dependency at version level.
- For each component, record the business owner, technical owner, data owner, decision owner and incident owner.
- Before purchase or integration, review evidence on security, privacy, data provenance, limitations and support.
- Minimise data and tool permissions; do not give the provider unnecessary access.
- Require notification and renewed testing for a version change, outage, data incident or material terms change.
- Preserve traces connecting prompts, outputs, model versions, tool calls and provider events.
- Prepare and test data portability, alternative service, rollback and safe-exit plans for provider change or loss.
How do you audit it?
- Are all third-party components and sub-dependencies known?
- Is an accountable human owner inside the organisation named for every risk and decision?
- Was a provider statement tested through independent or otherwise suitable evidence?
- Are data use, retention, subprocessors and limits recorded intelligibly?
- Can silent version or configuration changes be detected?
- Are record access, notification, cooperation and rollback terms workable during an incident?
- Was the exit plan tested against real data and dependencies?
Limit
The parties' legal status and obligations vary with country, sector, contract, data type and practical control. Public provider documentation may not disclose every sub-dependency or risk of the particular service you receive.
Remember in one sentence
You may use an external component; you cannot make accountability disappear.
Sources for this record
- S26NIST SP 800-53 Rev. 5, *Security and Privacy Controls*Standard
- S38NIST AI 600-1, *Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile*Voluntary standards-oriented institutional guidance
- S41NCSC and partner agencies, *Guidelines for Secure AI System Development*Official multi-agency security guidance
- S43NIST SP 800-161 Rev. 1 Update 1, *Cybersecurity Supply Chain Risk Management Practices for Systems and Organizations*Voluntary standards-oriented institutional guidance
- S44NIST, *AI Risk Management Framework Playbook* and AI RMF CoreVoluntary implementation guidance
- S47NIST, *Privacy Framework 1.0: A Tool for Improving Privacy through Enterprise Risk Management*Voluntary privacy-risk framework

