Australia Guidance for AI Adoption (2025)
Practice 5: Test and monitor – Australia Guidance for AI Adoption (2025)

Australia Guidance for AI Adoption (2025) 5.3.1: 5.3.1 Conduct capability-scaled safety evaluations

Developers of general-purpose AI should run safety evaluations that scale with model capability, for example assessing cyber-offensive capabilities and vulnerabilities, testing for chemical, biological, radiological and nuclear information risks, evaluating behaviour beyond intended uses, testing for jailbreaking and prompt manipulation, data privacy risks, or comprehensive red teaming.

Maintained by Gerard Blokdyk

Other controls in Practice 5: Test and monitor – Australia Guidance for AI Adoption (2025)

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.