Jailbreak Detection API for AI Products

Where it fits

• Screening prompt traffic before it reaches a high-privilege model or agent tool.
• Adding security telemetry to moderation, support, coding, or research assistants.
• Testing whether a known jailbreak family still works after prompt or model changes.

Operational steps

• Send the candidate prompt, recent turns, policy name, and app surface to the detection endpoint.
• Use the response severity to block, review, log, or downscope the request.
• Replay known jailbreak families in staging before release.
• Feed confirmed misses back into a custom test pack for future regression scans.

Common risks

• A single-turn classifier misses gradual roleplay or authority-shifting attacks.
• The app detects toxic language but not attempts to override system instructions.
• Detection logs contain sensitive user text without minimization controls.

How PromptGuard Scan fits the workflow

PromptGuard Scan combines a maintained jailbreak library with scan reports and API responses that fit product telemetry, CI checks, and security review workflows.