AI agent security
How tool access, delegated authority, memory, and external content change the attack surface of agentic systems.
Research
Practical research for emerging security problems.
Our research priorities sit at the intersection of AI behavior, software architecture, and real-world adversarial use. We are open to scoped collaboration with technical and research teams.
Adversarial evaluation
How evaluation design can produce reproducible evidence about misuse resistance and security-relevant behavior.
Application trust boundaries
How identity, authorization, APIs, and product logic interact to create vulnerabilities in modern platforms.
Security benchmark design
How narrowly defined, transparent benchmarks can support better engineering decisions without overstating coverage.
Collaboration formats
Defined around the research question.
01Scoped adversarial evaluations
02Benchmark and test-case development
03Joint vulnerability investigation
04Independent review of security assumptions
Related insights
Read the current research notes.
Start a conversation
Interested in a focused security research collaboration?
Tell us what you are building, the question you need answered, and where an independent security perspective could help.
Contact H-One