Research
Study the boundaries agents cross.
Attack taxonomies, source audits, and case studies help turn agent security questions into tests. Read the published methods and their limits.
01Published work
Methods you can inspect.
Attack taxonomy
A public vocabulary for AI attack vectors and their relationship to OWASP LLM Top 10 and MITRE ATLAS. Read the original publication and the maintained taxonomy.
Read the taxonomy researchEvolutionary Security Framework
A ten-phase model for hardening agentic AI systems, from identifying threats to testing defenses. It is a method for organizing security work.
Read the frameworkMCP trust boundaries
A source-code audit of the Python and TypeScript MCP SDKs examines boundary-crossing vulnerability classes and the conditions behind them.
Read the MCP audit02Disclosures
Report a security issue.
For a vulnerability in Tachyonic, contact security@tachyonic.co. Include the affected surface and a reproduction that avoids exposing other people's data. Test only systems you own or have permission to assess.