BoundaryBench: MCP Authority Drift Evals

Open-source deterministic tests for MCP agents that catch permission and authority drift before deployment.

Public Authorization-Boundary Benchmark for Tool-Using AI Agents

A synthetic public benchmark for testing whether agents act only with current identity, consent, authority, scope, and policy evidence.

Open benchmark for prompt-injection robustness of tool-using AI agents

An Independent Open-Source Safety Benchmark for Frontier LLMs and AI Agents

An independent, open, reproducible benchmark for safety-critical behaviors in frontier LLMs and tool-using AI agents.

Open-source behavioral containment benchmark for AI agents

Testing whether tool-using AI agents respect external behavioral contracts, with signed evidence when they fail.

New way of Thinking about Benchmarks

Exploring ways of Benchmarking that do not get saturated over time

GroundGate,external gate + benchmark for confident-but-wrong AI actions

Internal confidence fails at scale. A model-independent runtime gate + reproducible cross-vendor benchmark for confident-but-wrong AI actions.

A Local Testbed for AI Agent Security Failures

Reproducible tests for prompt injection, poisoned memory, unsafe tool use, and practical containment.