Agent systems
Identity, delegated authority, credentials, tool access and high-impact actions across an agent workflow.
ADVISORY / AI AGENT SECURITY
Find where agent authority can escape its intended boundary.
I review agent workflows, MCP servers and automation paths before they become difficult to unwind. The result is a technical assessment with evidence, priorities and controls your team can implement.
SCOPE / 01
The systems I review.
MCP environments
Server authentication, tool descriptions, parameter boundaries, cross-server influence and request-level evidence.
Software supply chain
Packages, install scripts, CLIs, skills and external components that agents discover or execute.
DELIVERABLES / 02
Evidence your team can act on.
This is not a compliance checklist or a generic AI review. The work follows the real execution path through identities, tools, packages and infrastructure.
- 01A reviewed attack-path map
- 02Prioritized findings with severity and evidence
- 03Control recommendations tied to the actual architecture
- 04A working session with engineering and security
GOOD FIT
Your agents already touch systems that matter.
The assessment is most useful when a team has a working architecture, is approaching production, or needs to understand a concrete exposure before scaling.
CONTACT
Tell me what you’re building.
Reach out about Oktsec, AI agent security, technical assessments, advisory or talks. I work with a small number of teams where the problem is concrete and the work can be useful.
OR EMAIL DIRECTLYgus@oktsec.com