ARENA / 01 / RUNNER

Pre-screen adversarial text before deeper testing.

This deterministic pre-screen matches phrases and nearby terms in the text you provide. It does not execute or observe a target model, agent, tool, or API.

ENGINE 2.1.0

01 PAYLOAD TEXT02 SIGNATURE SCAN03 PRE-SCREEN RESULT

ATTACK SIDE

Payload and probe surface

PAYLOAD.01

Adversarial payload0 / 10,000 characters / browser state with anonymous Firestore backup when available

Release profile

Attack packsJailbreakAttempts to override policy, role boundaries, or safety guardrails.built-in / v1.0.0 / Jailbreak03Harmful adviceRequests for weapons, self-harm, credential theft, or unsafe instructions.built-in / v1.0.0 / Harmful advice03Privacy leakageExtraction of secrets, personal data, credentials, or sensitive logs.built-in / v1.0.0 / Privacy leakage03Prompt injectionInstructions that target hidden prompts, system messages, or retrieved content.built-in / v1.0.0 / Prompt injection03Unsafe tool useActions that spend money, send messages, delete data, or execute commands unsafely.built-in / v1.0.0 / Unsafe tool use03

5 packs / 15 probes armedPayload required

DEFENSE SIDE

Release gate response

RUN.PENDING

PRE-SCREEN RESULTSTANDBY

SCORE--/100

SEVERITYNone

Awaiting payload

The result will show configured text matches, severity, and a concrete mitigation path.

Open scorecard