Infisical hits $600 million as startups pay to keep passwords away from AI agents

Infisical, a San Francisco security company, said on October 27 that it raised $40 million in Series B funding at a $600 million valuation, to build tools that let AI agents use company systems without holding passwords or API keys.

Elad Gil led the round. He also invested in Infisical’s Series A, and the company named Gradient and Y Combinator among its other backers. Infisical said the money will go to product development, hiring in engineering and go-to-market, and security infrastructure for AI agents.

Agent Vault keeps the keys out of the agent

The product at the center of the pitch is Agent Vault. According to Infisical, companies can define what each agent is allowed to reach, and the agent cannot leak credentials during an attack or send them by accident to LLM providers or other tools it talks to.

Infisical’s own documentation describes the mechanism more concretely. Agent Vault is an open-source HTTP proxy that sits between an agent and the APIs it calls, such as GitHub or Stripe. The agent sends traffic through the proxy with a time-limited session token, and the proxy attaches the real credential to approved requests as they leave the network. The idea, as the company explains it, is that prompt injection or a poisoned web page cannot make an agent reveal a secret it never had.

The same documentation says Agent Vault is a separate product from Infisical’s main secrets management platform. It runs on Infisical Cloud in the US and EU or can be self-hosted, and it is still in preview, so features and pricing may change. The funding announcement gives no pricing or general availability date.

“You wouldn’t give an employee unrestricted access to every system in your company, and AI agents shouldn’t be any different,”

said co-founder and CEO Vlad Matsiiako.

Customer numbers come from the company

Infisical said it serves more than 1,300 customers and 300,000 developers, protects 10 billion secrets a day, and that its open-source project has more than 30,000 GitHub stars. It listed Airbus, Volkswagen, Hugging Face, OpenRouter, Lucid, First American Financial Corporation, Wise and Clay as customers. None of these figures has been independently verified, and the release does not say how many customers use Agent Vault specifically, or how much revenue the company earns.

For teams already running Infisical for certificates, encryption keys or privileged access, the release positions agents as a new kind of user on the same platform. For everyone else, the open-source code means the proxy idea can be tested without a contract, though the preview label is a reason to keep it away from critical workloads for now.

The hard part is the proxy, not the vault

The technical angle is why this approach is different from storing secrets in a vault and reading them out at runtime. In the usual pattern, a program fetches a key and holds it in memory, and anything that can read that memory or the agent’s context can copy it. Agents widen that exposure because they ingest untrusted text, such as web pages and documents, and can be steered by it. Infisical’s design moves the credential out of the agent’s reach entirely: the agent only holds a session token that expires, and the proxy decides which calls get the real key.

That shifts the difficulty to policy. Someone has to decide, per job, which services an agent may call, which Infisical handles through what it calls access bundles in its documentation. The proxy also only controls traffic that passes through it, so an agent with other network paths, or one that is allowed to call a service and misuse that access, is a different problem. The release does not address those cases.

The timing fits a broader move toward running agents in isolated environments. Our earlier report on Supermicro’s Vera CPU rack for agent sandboxes covered the hardware side of the same concern, and investors are also backing agent software generally, as in the Slingshot seed round.

Gil framed the opportunity in broad terms, saying AI agents are becoming a new class of users inside every enterprise and that securing their access “will be one of the biggest cybersecurity challenges of the next decade.” That is an investor’s view, and the release offers no data on how many agent-related breaches have occurred or how often credentials leak from agents today.

What the round does and does not settle

The round shows that investors will pay a premium for a specific slice of agent security. What it does not show is how well Agent Vault performs against real attacks, how it compares with rival approaches, or when it leaves preview. Those are the points to watch as Infisical spends the new money.