UK AI Cyber Security Code of Practice 9.4: Evaluate outputs for leakage of model internals and unintended influence

What does 9.4 require?

Developers should evaluate model outputs to confirm they do not let system operators or end-users work out non-public details of the model or of its training data, and that they do not hand operators or end-users unintended influence over the system.

Maintained by Gerard Blokdyk

Other controls in Secure development (Principles 5 to 9) – UK AI Cyber Security Code of Practice

5.1 Comprehensive asset inventory including interdependencies
5.2 Processes and tools to track, authenticate, version and secure AI assets
5.3 Disaster recovery plans tailored to attacks on AI systems
5.4 Protect sensitive data and sanitise inputs, with proportionate protection of weights
6.1 Evaluate access control frameworks for APIs, models, data and pipelines
6.2 Controls on externally offered APIs against attacks on the AI system
6.3 Dedicated, segregated environments for development and model tuning
6.4 Publish a clear vulnerability disclosure policy

Query this from an agent

The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.

MCP endpoint All UK AI Cyber Security Code of Practice controls