Island's $400M bet: the browser is your agent control point
Island raised a $400M Series F at $6.4B selling an enterprise browser that inventories AI agents, maps their MCP servers and trims their permissions. The control point moved.
Island raised a $400 million Series F on September 24 at a $6.4 billion valuation, and the pitch it sold investors is worth more to you than the number. Island ships a Chromium-based enterprise browser, and it now describes itself as "the agentic control plane for enterprises." Not an endpoint agent. Not a network appliance. The browser. That is a claim about where AI agents actually do their damage, and it is correct.
What actually happened
Per Island's announcement, Evolution Equity Partners led the round, joined by Sequoia, Coatue, Insight Partners, Cyberstarts, J.P. Morgan Growth Equity Partners, Georgian, Prysm, G Squared, Alta Park, Squarepoint, and a personal check from CrowdStrike co-founder Dmitri Alperovitch. The company says the $6.4 billion post-money is more than double its 2024 valuation, that it employs about 1,000 people, and that it has doubled ARR every fiscal year since launching in 2022. It did not disclose a revenue figure, so treat the growth-rate claim as company-reported.
The agent feature list is the interesting part. Per SiliconANGLE, the product builds an inventory of the AI agents employees actually use, maps the MCP servers, files and extensions those agents touch, blocks high-risk tools, filters malicious prompts, strips over-broad permissions, and logs inference spend so you can see which tools nobody uses. SiliconANGLE also reports plans to add roughly 500 people by mid-2027.
Read that list again. Half of it is asset management. The security features are downstream of simply knowing what is running.
Why the browser matters for your business
The agent runs inside your session, with your cookies. When an employee points a browser agent at your Shopify admin, your bank portal or your CRM, it is not authenticating as a service account with a scoped token. It is riding a human's logged-in session, with that human's full permissions, in a tab nobody is watching. Every access-control decision you made at the API layer is bypassed by a thing that clicks buttons. That is why a browser company just raised $400 million.
You will not buy Island. You still have the problem. A ten-person shop does not deploy an enterprise browser fleet. But the three cheap moves are the same ones Island productized: know which agents your team runs, know what credentials those agents inherit, and keep a log of what they did. We made the counting argument last week and it keeps getting cheaper to ignore and more expensive to have ignored.
Permissions rot faster than agents do. Island ships permission-trimming because broad scopes accumulate silently — the same failure mode behind agents inheriting dead access. An agent granted write access for one migration in March still has it in September. Put an expiry on every scope you hand an agent, or you are running a permanent standing privilege you cannot name.
Inference logging is a budget control, not a security feature. The fact that a security vendor bundles cost visibility tells you what its buyers keep asking for. If you cannot answer "which agent spent what, on which model, last month," you do not have an AI budget. You have a bill.
Key takeaways
- Island raised $400M Series F at a $6.4B valuation, led by Evolution Equity Partners, announced September 24
- The company now positions the browser as "the agentic control plane" — not the endpoint, not the network
- Features include agent inventory, MCP server mapping, permission trimming, prompt filtering and inference cost logging
- ARR growth and the 2024 valuation comparison are company-reported; no revenue figure was disclosed
- Browser agents ride a logged-in human session, so your API-layer access controls do not apply to them
- Without the enterprise product, do the three cheap things: inventory agents, scope their credentials, log their actions
If you cannot name every agent with access to your admin panel, that is the project. We map what your team is running, scope each agent to the minimum credential that does the job, and wire an audit log you can actually read. See how we govern agent access, or ask for an agent and permissions inventory.
Sources: Island, SiliconANGLE.
- #ai-agents
- #browser-security
- #agent-governance
- #mcp
- #funding
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Strada browser automation: agents for portals with no API
Strada shipped browser automation on September 24 so AI agents can drive carrier portals that have no API — recorded once, run on live data, logged end to end.
Read itPrismML's 1-bit Bonsai model runs a VLM on smart glasses
PrismML showed a 2B vision-language model at 1-bit precision on Snapdragon AR1, using 4x less memory and running 2x faster. On-device inference just got cheaper.
Read it