Island Bets $6.4B That the Browser Is the Agent Control Plane
By @sharedot · · 6 pages
- AI Frontier
- Agent Security
- Enterprise Browsers
- Agent Governance
Island raised $400M at a $6.4B valuation to make its enterprise browser the point where AI agent actions are blocked before execution.
What Happened: A $400M Round for the 'Agentic Control Plane'
Island, the enterprise browser company, has secured $400 million in Series F funding at a $6.4 billion valuation, led by Evolution Equity Partners with participation from Sequoia, Coatue, Cyberstarts, Insight, J.P. Morgan, Georgian, G Squared and others.
CrowdStrike co-founder Dmitri Alperovitch also joined as an investor. According to Forkast News, the round — a 33% step-up from the company's $4.8 billion Series E in March 2025 — brings total funding past $900 million, with ARR reportedly doubling every fiscal year since the 2022 launch.
The capital is earmarked to position the Island Enterprise Browser as the control plane governing both human and AI agent activity inside enterprises.
Why It's Surprising: Enforce at the Action, Not After It
The bet lands at the same moment the agent-security conversation is shifting from monitoring to pre-execution enforcement.
Writing in Cybersecurity Insiders, Michael Kantor of Hashgraph Online argues that a trace showing an agent just deleted a file is not a control — the real security boundary is the moment between an agent requesting an action and the tool executing it.
Island's architecture takes that logic into the browser: by owning the environment where agents browse and reach SaaS apps, every agent activity must pass through a single governed point, enabling granular control over data exfiltration and credential surrogation.
CEO Mike Fey said the browser is 'the only place where you can govern the intersection of humans, agents, and data at scale.'
The Evidence: A Converging Agent Governance Stack
Forkast News frames Island's step-up as market validation that browser-native control is a leading contender for the enterprise standard, alongside kernel-level enforcement (Meta Sentinel) and persistent cloud VMs (GrokBot), and distinct from API-level monitoring vendors like Dataiku and SAP.
Cybersecurity Insiders points to regulatory tailwinds: NIST's AI agent security work, including its Request for Information on securing AI agent systems, and OWASP's 2026 GenAI Top 10 update, which treats excessive agency as a core risk.
The practitioner playbook from Kantor's piece — least privilege at the tool layer, separating reversible from irreversible actions, treating external content as untrusted input, and preserving an auditable decision trail — maps directly onto what a browser-level control plane can enforce.
The Stakes: Coverage, Not Demos
The hard part, per the Cybersecurity Insiders piece, is coverage: a policy engine can look impressive in a demo while an agent reaches the same side effect through an unmonitored path like an MCP config change, a plugin install, or an external tool call.
Teams should explicitly document which actions are blocked, which need review, and which sit outside enforcement — and design policy so routine work stays fast while genuinely sensitive actions gain friction, otherwise users learn to approve everything.
Forkast News concludes the agent economy is now about the control plane, and whether it lives in the kernel, the VM, or the browser will determine how enterprises manage autonomous-agent risk.
A practical starting point: inventory agent actions, classify them by consequence, define hard stops for destructive operations, and place controls as close to execution as possible.