privacy-testing

AgentLeak (IEEE Access 2026): benchmark for internal-channel privacy leakage in multi-agent LLM systems. Inter-agent messages leak 68.8% vs 27.2% in final outputs; output-only audits miss 41.7%. 1,000 scenarios, 7 channels, 5 LLMs. Tool & MCP server: github.com/yagobski/agentleak

Open-source privacy testing for AI agents. Python SDK, CLI and MCP tools to detect data leaks across tool calls, memory, inter-agent messages and logs, with redacted reports and CI gates.

OutageFence by Slanta Cause — free local tests for data boundaries when dependencies fail.

€250 public TypeScript/React repairs and ScanBell consent-regression QA by Timothy Peterson

Privacy CI for Zcash applications: deterministic behavior tests, inspectable evidence, baseline comparisons, and CI gates. Built for ZECATHON.

Automated Playwright harness that verifies PixPass does not upload photo data during core user flows. Run it yourself against production.

Test blur, redaction, masking, and anonymization on every image/video frame. Deterministic PASS/FAIL, temporal tracking evidence, local reports, and CI-ready exit codes.