About
About Ultra13
Ultra13 builds context-firewall controls for AI agents that read untrusted context and take real actions.
Why Ultra13 exists
AI agents collapse data and instructions into the same execution loop. Once agents use RAG, memory, MCP, tools, browsers, APIs, and customer data, the security boundary is no longer the prompt. It is the path from context to action.
What we build
Ultra13 maps, attacks, controls, and continuously validates agent workflows with source-to-sink policy — enforcing which context sources may influence which actions across memory, retrieval, MCP, tools, egress, and approvals.
Who we work with
- AI startups approaching production or enterprise review
- CTOs and engineering teams shipping agents with real permissions
- Security teams governing AI agents
- Agent platforms selling into enterprise buyers
Operating principles
- Evidence before claims
- Context provenance before model vibes
- Enforcement before prompt-only advice
- Runtime controls before static PDFs
- Replayable validation after every change
Team
Team bios are being finalised. If you are evaluating Ultra13 for a security review and want to speak with the team, join the waitlist and we’ll reach out.