AI Risk Classification

Risk classification is used to decide how much evaluation, red teaming, and security review an AI agent needs before release.

Data requirements

• Evaluation suites (text) — Contain the intended behavior and security requirements used for classification and testing.

AI methods and techniques

• Evaluation suites — Used to encode intended agent behavior and security requirements.
• Risk-calibrated release checkpoints — Used to decide what level of review an agent needs before release.
• Red teaming — Applied more heavily to higher-risk agents.
• Security review — Added as part of the release process for higher-risk agents.

Industries

Real-world evidence

1 documented case study on record.

Companies using this: Expedia.