Open benchmark · v1

Standardized adversarial
benchmarking for AI systems

Submit your model endpoint. Cybertope runs a standardized suite of prompt injection and jailbreak tests, scores the results, and publishes your model's security posture to a public leaderboard.

OWASP LLM01

Prompt Injection

5 test cases covering instruction override, role reassignment, delimiter injection, context exhaustion, and system prompt extraction.

OWASP LLM07

Jailbreak & Alignment Bypass

5 test cases covering persona adoption, hypothetical framing, encoding obfuscation, many-shot bypass, and competing objectives.