Cybersecurity Agent

AI security agent that audits MCP skills and agent code for vulnerabilities. Detects prompt injection, tool poisoning, and credential leaks. Issues cryptographic trust certificates.

AI cybersecurity agent — Solidity exploit, root cause analysis, threat detection

Agent Trust Arena is a security benchmark for evaluating AI agents' ability to establish trust, detect threats, and maintain secure collaboration in multi-agent enterprise workflows.

This green agent evaluates the behavior of an AI system by simulating benign but potentially sensitive usage patterns and telemetry events. It assesses how the target system detects, classifies, and responds to these scenarios, including risk scoring, policy enforcement, and response consistency. The agent focuses on end-to-end evaluation of safety handling, signal interpretation, and decision outcomes, ensuring the system operates within expected guardrails without manual intervention.

Open-source red-teaming framework for AI agent systems. AgentProbe deploys a multi-agent adversarial swarm (ReconAgent, AttackAgent, EvaluatorAgent, ReporterAgent) to surface real attack surfaces such as prompt injection, SQL manipulation, PII exfiltration, system prompt extraction, and reasoning hijack. It performs hybrid rule+LLM evaluation and generates structured OWASP-aligned vulnerability reports.