The short answer
Investors are betting that the hard part of enterprise AI agents is control, not capability. Island started as a secure enterprise browser. It now describes itself as an “agentic control plane” that decides what employees and AI agents can reach, what they can do there, and how every action gets logged. A $400 million round at more than double its 2024 valuation says buyers are ready to pay for that layer.
If you are putting agents onto CRM, ERP, code or payment systems, the same questions now come up in security review whether you buy a platform or not: which identity does the agent use, what can it touch, and who can prove what it did. They are easier to answer when you design them into AI agent development from the start than when you bolt them on before launch.
What did Island announce?
Dallas-based Island announced a $400 million Series F on Thursday, September 24. Evolution Equity Partners led the round, and most of Island’s existing backers came back in. The company says its annual recurring revenue has doubled every fiscal year since its 2022 launch. It did not disclose revenue figures. CEO and co-founder Mike Fey told CNBC the money will go into research and development and into expansion across Europe, Asia and the Middle East, with headcount growing from 1,000 to 1,500 by the middle of next year.
The more important part is the change in positioning. Island built its business on an enterprise browser that lets security teams control what employees do inside web apps. The new pitch goes further: one control plane that governs people and AI agents across devices, browsers, applications, networks and data. Island lists five layers for it: last-mile control, network, data, identity and observability. The goal is real-time visibility into who or what is acting, what it can access, and whether each action should be allowed.
Why is AI agent governance attracting this much money?
Agents have moved from demos to accounts with real permissions. They read mailboxes, update CRM records, open pull requests and trigger payments, often through a shared service account or an employee’s own session. Most existing security controls were designed for a person at a keyboard. “Every old control is breaking, so everything’s up for grabs,” Fey told CNBC. CNBC placed the round in a wider scramble to defend businesses against rogue AI agents.
The competition is crowded. Island faces incumbents such as Palo Alto Networks and a long list of startups chasing the same AI security budget. This year has also brought deals for agent identity and gateway tooling across the market. For buyers the signal is clear: agent governance is becoming its own product category, and auditors and customers will start asking about it in the same way they ask about SSO and encryption today.
What it means for US & EU software teams
First, agent identity is now a design decision, not an afterthought. An agent that acts through a developer’s token or a shared admin account makes every action look like a human did it. When something goes wrong, nobody can tell which agent acted, on whose request, or why. Give each agent its own service identity with short-lived, narrowly scoped credentials, and bind every tool call to that identity and the originating request.
Second, the audit trail is what makes an agent shippable. Security reviews for SOC 2, ISO 27001 and customer questionnaires increasingly ask how automated actions are authorized and logged. In the EU, the AI Act’s logging and human-oversight duties for high-risk systems, and DORA for financial entities, push the same way. A log that records the prompt, the tool called, the parameters and the result for every agent step answers most of those questions. Without it, agents tend to stall in pilot.
Third, buying a control plane does not replace secure agent design. A platform like Island’s can watch and block actions at the edge. It cannot fix an agent that trusts every document it reads or that has write access it never needed. Prompt injection, over-broad tool scopes and missing approval steps are architecture problems. They are best caught by testing the agent itself, including adversarial testing as part of penetration testing and security audits, before it goes live.
What to do now
- Inventory your agents. List every agent, copilot and automation that acts on business systems, the credentials it uses, and what it can read or change.
- Separate identities. Replace shared accounts and borrowed user tokens with one identity per agent, short-lived credentials and least-privilege scopes.
- Gate risky actions. Require human approval for payments, deletions, permission changes and anything sent outside the company.
- Log every step. Record prompt, tool, parameters, result and requesting user for each agent action, and keep it where your security team already looks.
- Test before launch. Run prompt-injection and privilege-escalation tests against agents the way you test any other internet-facing app.
Frequently asked questions
What did Island announce?
On September 24, 2026, Island announced a $400 million Series F led by Evolution Equity Partners at a $6.4 billion valuation. Existing investors including Sequoia, Coatue, Insight Partners and J.P. Morgan Growth Equity Partners took part. The company says the round brings total funding to about $1 billion and that its valuation has more than doubled since 2024.
What is an agentic control plane?
It is a single layer that decides what people and AI agents may do across an organization’s systems. In Island’s description it covers identity and access controls for agents, guardrails, cost governance and a full audit trail across browsers, endpoints, cloud and internal resources, so a security team can see who or what is acting, what it can reach, and whether an action should be allowed.
Why are investors paying for AI agent security now?
Companies are giving agents real access to business systems faster than their controls can keep up. CNBC tied the round to a wider scramble to defend against rogue AI agents, and Island CEO Mike Fey told CNBC that every old control is breaking. The money follows the gap between how quickly agents are being deployed and how few teams can govern them.
Do we need an enterprise browser to govern AI agents?
No. A vendor control plane is one option. The same goals can be met in your own architecture: give each agent its own identity, scope its credentials to the minimum, route its actions through a policy check, and log every call. The right choice depends on how many agents you run, where they act, and what your auditors will ask for.
What should teams building AI agents do first?
Inventory every agent and the systems it can touch, then give each one its own identity and least-privilege credentials instead of a shared service account. Add human approval for irreversible actions, keep an audit log that ties each action to an agent and a request, and test agents for prompt injection before they reach production.
Sources
Island — Island Announces $400 Million Series F, Bringing Valuation to $6.4 Billion
CNBC — Cyber startup Island hits $6.4 billion valuation in new round as AI attacks fuel spending wave
SecurityWeek — Island Raises $400 Million at $6.4 Billion Valuation
SiliconANGLE — Enterprise browser developer Island raises $400M at $6.4B valuation