Back to Ecosystem Pulse
EvaluationProduction Ready
safelabs-eval
by AgentSafeLabs
OWASP ASI-aligned red-teaming and evaluation for AI agents
Python
Updated Oct 25, 2026
Share:
Summary
Run OWASP ASI-aligned red-teaming and evaluation campaigns against AI agents. Implements a test harness and attack templates for prompt-injection, jailbreaks, and agent-specific failure modes, with tooling to automate repeatable adversarial runs. Includes integrations and example workflows for popular agent stacks so teams can plug tests into CI and developer workflows. prompt-injection
Key Benefits
As agents become autonomous and delegate tasks, security failures propagate quickly across systems; traditional single-LLM tests miss agent-to-agent interaction flaws. This framework treats adversarial evaluation as first-class, surfacing systematic jailbreaks, delegation failures, and interaction-level risks that degrade trust. By providing standardized, OWASP-aligned tests, it makes agent track records comparable and repeatable across teams. OWASP-aligned tests
When to Use
Security engineers and AI teams running pre-production adversarial tests and continuous agent evaluation pipelines. Evaluation-Driven Development
Applications
- When you need repeatable red-team campaigns against multi-agent workflows
- Detect prompt-injection and jailbreak patterns in agent delegation
- Integrate adversarial tests into CI to track agent reliability over time
Works With
langchaincrewai
Topics
agent-securityai-safetycrewaievaluationjailbreaklangchainllm-securityowaspprompt-injectionred-teaming
Similar Tools
agent-playground
Keywords
multi-agent trusta2a evaluationred-teamingagent-security