AI Agents Force a New Security Layer for Enterprise Teams

AI agents are moving from chat windows into the systems where companies store sensitive data, approve work, and execute business processes. That shift is creating a security problem that traditional identity tools were not designed to solve.

What happened

Cymphony, a New York- and Tel Aviv-based startup, raised $30 million in new funding, including a $25 million Series A co-led by Sequoia Capital and SMBC Fin Atlas Beyond Fund, according to TechCrunch. The round values the company at more than $100 million.

Cymphony is building a “workforce graph” that combines identity, data, and activity signals across employees, AI agents, and other non-human identities. The company says it found roughly 85,000 files exposed to AI tools at one public company, although it reported that those files were not accessed through the systems it investigated.

The startup also described an incident involving an unsanctioned Claude installation that used an external collaborator’s existing access to scan thousands of sensitive files. Its platform can investigate incidents, prioritize remediation, and correct access permissions, with customers choosing how much automation security teams want to allow.

Why it matters for builders

The important change is not simply that agents can call more tools. Agents can operate at machine speed, change their route through a workflow, acquire capabilities at runtime, and sometimes create additional agents. A permission model based only on named human users therefore misses the actor that actually touched the data.

For teams building automations, the practical response is to treat every agent as a first-class identity. Record which workflow version acted, which tool it invoked, what data it read, and what policy authorized the action. Separate read, write, and external-send permissions, then add approval gates for irreversible steps.

Cymphony’s funding is a market signal, but the engineering lesson is more immediate: agent deployments need an identity and activity layer before they need more autonomy. The safest agent is not the one with the fewest capabilities. It is the one whose capabilities are explicit, observable, and revocable.