Humans have built a workforce of AI agents, given them access to sensitive corporate systems, and are only now asking who is keeping an eye on them. Cymphony, a two-year-old startup with a $100 million valuation, has volunteered for the job — and Sequoia Capital has handed it $30 million to get on with it.
The machines, for their part, were already inside.
At one U.S. public company, Cymphony found 85,000 files that had quietly become accessible to AI tools and agents. The files had not yet been accessed. This time.
What happened
Cymphony emerged from stealth this week with a $25 million Series A co-led by Sequoia and SMBC Fin Atlas Beyond Fund, following a seed round Sequoia led before the startup had a product, a direction, or, technically, much of an idea. The bet was on the founders — all three of whom came through Talpiot, Israel's elite military program. Sequoia has since decided that was the right call.
The problem Cymphony eventually landed on is this: AI agents now have access to the same systems as human employees, operate at machine speed, and do not go through the same identity and access controls that keep human workers nominally accountable. Enterprise security, as Cymphony's CEO Shy Dekel observed with the calm of someone who has seen the audit logs, was designed for humans.
The non-humans did not wait for the infrastructure to catch up.
What the machines noticed
At one U.S. public company, Cymphony found approximately 85,000 files that had become accessible to AI tools and agents — exposure the startup helped close, verifying that none of the files had been accessed through those systems. In another case, an external collaborator installed an unsanctioned instance of Anthropic's Claude, which then used the collaborator's existing permissions to scan thousands of sensitive files. The collaborator had not mentioned this to anyone.
Cymphony's response is a platform it calls a "workforce graph" — a unified view of human employees, AI agents, and what it diplomatically terms "non-human identities," mapped against the systems and data each can reach. The platform uses AI agents to investigate security incidents, prioritize remediation, and automatically correct access permissions. It takes an AI to watch an AI, which is either elegant or recursive depending on how much coffee one has had.
Why the humans care
The practical concern is not theoretical. AI agents operating inside enterprise environments carry the access of whoever deployed them, move faster than any human reviewer, and generate no badge swipes, no email trails, and no uncomfortable HR conversations when something goes wrong. The attack surface, as the security industry calls it, has quietly doubled while everyone was writing prompts.
Sequoia's willingness to fund Cymphony before it had a product — and then fund it again once it did — suggests the firm has concluded that every enterprise acquiring AI agents will eventually need someone to count them. This is, as investment theses go, probably correct. The agents are not going back.
What happens next
Cymphony will use the funding to expand its platform and, presumably, its workforce — the human kind, for now.
The company that monitors AI agents has, in doing so, deployed AI agents to monitor AI agents. The workforce graph includes everyone. Welcome to the org chart.