Ultra Security, a startup building runtime controls for AI agents, announced a $10 million seed round on October 21, 2026, as it emerged from stealth. The company's release names FDVC as an investor, which Caplight expands to Forward Deployed Venture Capital; the research did not establish whether the release identifies a lead investor, a valuation or a full investor list.
What Ultra Sells
According to its About page, Ultra positions itself around runtime enforcement rather than model development or agent orchestration. Those are the company's own descriptions, not independent findings.
The product connects to local and remote MCP servers. MCP, the Model Context Protocol, is an open standard Anthropic introduced on November 25, 2024 to link AI applications to tools and data. Ultra's documentation describes request and response tracing, logging, auditing and measurement.
Not everything happens in real time. The documentation shows that the governance guardrails and the scheduled anomaly analysis operate differently from one another.
Then there is the question of where data goes. Ultra markets a local-first architecture, yet its anomaly documentation also describes activity synchronization to Ultra Hub. The research did not confirm which data leaves customer devices under each deployment setup, a gap that prospective buyers would presumably want closed.
The Founder
Chase Lee previously created Trustpage, which Vanta acquired in a deal announced January 19, 2023. Vanta's announcement said Trustpage had nearly 400 customers at that point. That figure belongs to Trustpage, not Ultra.
A post displayed on the LinkedIn profile of Saul Levin, of Entrée Capital, said the firm first met Lee in 2020 through an introduction from Jonathon Triest, and later backed him at Trustpage. The retrieved page labeled the post as shared content, and its original author was not established. Entrée Capital's company account also posted about Ultra in late October.
Lee has also posted on LinkedIn about leaving Vanta. The research did not establish a date for that post.
Customers and Pricing
Ultra's homepage carries a testimonial from a customer named Batchelor, who described making SharePoint, Outlook, Slack and Granola available to AI workflows across Martec. Because Ultra hosts the testimonial, it should not be read as an audited deployment result.
Pulse 2.0 reported that the launch material refers to customers and design partners but names no full roster and does not separate paying customers from evaluators.
The pricing page estimates Ultra's cost at roughly 1% to 3% of agent spend. That is a vendor estimate, not measured customer economics. The page also advertises FedRAMP-aligned enterprise deployment options, though the research did not verify any FedRAMP authorization.
The company describes itself as remote-first and did not confirm a headcount in the material reviewed. Stealth Startup Spy, in an October 29 newsletter, reported that Ultra spent 10 months in stealth. The research treats that figure as approximate.
A Crowded Field Under Rising Pressure
Ultra calls itself the first purpose-built product in its category. That is marketing language, and the research does not establish who came first. Other vendors, Operant AI among them, publicly advertise overlapping MCP and agent-runtime controls.
Standards bodies are drifting toward the same territory. In a September 2026 announcement, OWASP's GenAI Security Project unveiled an updated LLM Top 10 and an Agent Control Standard, extending its guidance toward runtime enforcement.
Survey data hints at the scale of the visibility problem. An April 21, 2026 Cloud Security Alliance release carried the headline that 82% of enterprises have unknown AI agents in their environments. It is a dated survey, however, not a current estimate of prevalence.
What's Next
FinSMEs coverage, reproduced by CB Insights on October 22, said Ultra plans to expand operations and development efforts. The company's announcement offered no hiring target, runway or spending timetable, so how far $10 million stretches remains an open question.
