The first generation of enterprise artificial intelligence lived mostly inside a chat window. The second is gaining passwords, tokens, access to databases, code repositories, and operational tools. It is at this juncture that AI stops being merely software that responds and becomes an entity capable of taking action within the enterprise. Cymphony, a startup based between New York and Tel Aviv, has raised $30 million to build the security layer needed to govern this new population of agents.
The funding includes a $25 million Series A co-led by Sequoia Capital and SMBC Fin Atlas Beyond Fund, following an earlier seed round by Sequoia. According to TechCrunch, the round brings the company's post-money valuation to over $100 million.
A digital identity that is not an employee
Companies have spent decades building identity and access management systems around people: named accounts, roles, authentication, approvals, revocations, and activity logs. An AI agent breaks part of this architecture. It can use service credentials, chain together different tools, and make thousands of decisions at speeds incompatible with manual supervision.
The risk does not necessarily require a malicious model. All it takes is excessive permissions, a manipulated prompt, a compromised dependency, or a misinterpreted objective for an agent to quickly do something that would take an employee hours to do.
Security must follow the action, not just the login
This is why the problem cannot be solved simply by asking the agent to authenticate. It is necessary to know which tools it can use, what data it can read, which actions it can perform, and in what context. A control system must also reconstruct the decision chain when something goes wrong.
Cymphony fits into a new market that aims to give enterprises visibility into already deployed agents, their identities, and associated privileges. It is an updated version of the principle of least privilege: an agent should only have the access necessary for the task it is performing, for the strictly necessary amount of time.
The paradox of speed
Automation produces value precisely because it eliminates waiting times and manual steps, but that same speed amplifies errors. A compromised human account can cause enormous damage; a compromised agent can also automate the damage. Cybersecurity in the agentic era will therefore have to shift from simple perimeter protection to continuous behavioral surveillance.
This means distinguishing between normal and anomalous actions, interrupting risky sequences, and maintaining readable audit trails. Governance becomes a component of the product, not a control tacked on at the end.
Sequoia is betting on a new category
The new investment is also interesting for what it says about the market. While billions pour into model makers and agents that write code or automate offices, a second economy is growing, made up of the infrastructure needed to make them usable in large organizations.
It happened with the cloud: after the explosion of platforms, companies focused on observability, identity, and security grew. AI agents could repeat the pattern on a much more compressed timeline.
The real test will come when agents become invisible
Today, a company can still count how many agentic projects it has launched. In the future, many agents could be embedded inside already deployed software and appear without a traditional provisioning process. The ability to discover them will become just as important as controlling them.
Cymphony’s thesis is therefore broader than a single security product: if enterprises are truly building a digital workforce, they will need to know who, or what, is operating within their systems. The next corporate organizational chart could contain thousands of identities that do not have a desk, but still possess privileges.
