Lesson 5 of 6
Policy and legal checks: SAP API Policy, data protection and provider terms
9 min
Listen to this lesson
Security controls answer "can we do this safely?". Contracts and law answer "may we do this at all?". This lesson lists the questions Nadia takes to legal counsel and procurement. It teaches patterns and questions, not legal conclusions; your counsel decides how they apply to your contracts.
The SAP API Policy
SAP's API Policy (version v.4.2026a at the time of writing) is part of the documentation for SAP solutions. Two sections matter for coding agents. Section 1.2 says customer and third-party applications must not use APIs that aren't published, except as the documentation permits or SAP authorizes; it gives custom-developed ABAP interfaces in private cloud and on-premise deployments as an example of permitted use. Section 2.2.2 prohibits API use for interaction or integration with (semi-)autonomous or generative AI systems that plan, select or execute sequences of API calls, except through and within the limits of SAP-endorsed architectures, data services or pathways expressly intended for such purposes. Section 3 says SAP may monitor usage and enforce, and that controls must not be circumvented through intermediaries.
- Is the agent using an SAP-shipped, documented pathway (for example SAP's ADT MCP server, licensed Joule for Developers)?Yes → record the documentation that describes it · No → continue
- Does the agent call ABAP or OData APIs that SAP publishes on the Business Accelerator Hub or in product docs?Yes → check the API's documented use and controls · No → continue
- Does it use non-published interfaces (for example ADT REST endpoints through a community MCP server)?Yes → take it to counsel; section 1.2 and 2.2.2 questions apply · No → continue
- Could it extract data at scale?Yes → section 2.2.2(b) on systematic extraction applies; redesign · No → document the assessment
SAP documents its ADT MCP server for use with MCP-capable coding assistants and says support for its MCP tools requires a valid Joule for Developers, ABAP AI capabilities license. Community ADT MCP servers are open-source projects that SAP doesn't support. Whether a particular setup is within the policy is for counsel; what the technical team owes counsel is an accurate description of which path is used, which interfaces are called and how much data moves.
Data protection
- What leaves the system: source code, object names, ATC results, and anything the agent reads: data preview results, short dumps (which can contain business data), test data and ticket texts.
- Where it goes: to the model provider under your contract, and to logs on the laptop and in your log store.
- Personal data: keep it out of development clients where possible; mask copies; restrict table display in the agent role; deny the feed tool so dumps stay out of context (lesson 3).
- Legal basis and processor terms: a data processing agreement with the provider, transfer mechanisms if data leaves the EU, records of processing.
- AI regulation: coding assistance is unlikely to be a high-risk use under the EU AI Act, but the Act's AI literacy duties already apply to organizations that deploy AI systems; train the people who use agents.
Provider terms: Claude Code as an example
| Question | What Anthropic documents for commercial use (October 2026) |
|---|---|
| Training on our code? | Anthropic doesn't train generative models on code or prompts sent under commercial terms unless the customer opts in, for example through its Development Partner Program |
| Retention | Standard 30 days for Team, Enterprise and API; zero data retention available to qualified Claude for Enterprise accounts, enabled per organization |
| Local copies | Session transcripts stored in plaintext under ~/.claude/projects for 30 days by default (cleanupPeriodDays) |
| Third-party clouds | Claude Code can run against Amazon Bedrock, Google Cloud and Microsoft Foundry, with those platforms' encryption and terms |
| Non-essential traffic | Metrics and error reporting can be disabled; on third-party clouds they are off by default |
| Feedback reports | /feedback (and /bug, /share) sends the conversation, including code, to Anthropic, where it is kept for 5 years; DISABLE_FEEDBACK_COMMAND=1 turns it off |
Data questions answered from Anthropic's Claude Code data usage documentation; recheck before contract signature.
The local transcripts deserve a line in the policy. They hold whatever the agent saw, in plain text, on a laptop. Contoso shortens the retention period in managed settings and relies on disk encryption on developer machines. For SAP's own tools the same questions apply in SAP's terms: Joule for Developers runs on SAP's AI services under SAP's contract, and SAP's Build Code docs remind users that if they bring their own model key, provider terms and data handling become their responsibility.
Questions for procurement and counsel
- Which SAP-documented pathways do we use, and do we hold the licenses SAP names for support (for example Joule for Developers for the ADT MCP tools)?
- Do any agents call interfaces that aren't published APIs? If so, under which permission in our SAP contract?
- What does our model provider contract say about training, retention, sub-processors and data location, and does it match the data the agent will see?
- Who is the data controller for logs and transcripts that contain code or business data, and how long do we keep them?
- How will we hear about changes? SAP's policy and the agent's terms both change; name an owner who checks them each quarter.
What would you do?
A developer wants to use a community ADT MCP server instead of SAP's, because it can also read table contents directly. What should Nadia do?
- API Policy v.4.2026a: section 1.2 on non-published APIs and 2.2.2 on AI systems that plan and execute API calls, except via SAP-endorsed pathways.
- Give counsel facts: which path, which interfaces, how much data; counsel gives the conclusion.
- Data protection: know what the agent reads, where it goes and how long it stays, including local transcripts.
- Check provider terms on training, retention, zero data retention and hosting before rollout.
Check your understanding · not graded
Lesson check
1. What does section 2.2.2 of SAP's API Policy v.4.2026a address?
2. What does section 1.2 of the API Policy give as an example of permitted use?
3. What is Anthropic's documented standard retention for commercial Claude Code users?
4. Why do short dumps matter for data protection in agent work?