Island Enterprise Browser for Agents Governs AI Browsing

Island introduced the Island Enterprise Browser for Agents on October 29, 2026, a new product that gives AI agents a governed way to access, read, and act on software-as-a-service applications, internal systems, and public websites. According to the company announcement, the browser applies the same enterprise controls and policies to agents that organizations already apply to human workers, rather than bolting safeguards onto consumer browsing tools.

The launch lands as agents take on routine enterprise work. According to Island, sales agents update customer records after calls, human resources agents help onboard new hires, and finance agents process invoices. Each of those jobs requires a browser to fill in forms, pull data, and chain actions across applications. The company argues that the browser has become the missing piece of enterprise agent infrastructure, and that leaving it unmanaged creates a second, largely invisible workforce inside the company.

Island is positioning the product as an agentic control plane for enterprises, a system for governing people and AI agents as they work across devices, browsers, applications, networks, and data. The company says millions of users already trust its enterprise browser, and the agent version extends that policy framework to automated workers.

Why agents need a governed browser

Until now, many organizations have relied on ad-hoc approaches that layer controls on top of consumer infrastructure, according to the announcement. That setup can expose credentials, let data escape from sanctioned applications, and leave records open to deletion. A browser built for agents changes the risk profile because the agent often holds powerful credentials and can move quickly across systems, repeating an error at machine speed before a human notices.

The core idea behind the Island launch is that organizations should govern browser activities, not agents in the abstract. With Island, organizations deploy AI products in the same way they deploy software today. The company says it governs browser activity from the moment an endpoint or cloud agent calls the browser or accesses internal resources, including data movement, credential usage, and resulting outputs, all according to the policies the organization has configured.

That framing matters because agent deployments span local machines and cloud infrastructure. Island says the product deploys on devices and in the cloud, covering local and remote agents with no dependency on a cloud-hosted browser outside the control of the organization. For regulated teams, keeping the browser inside enterprise control is a central selling point.

What is built into the engine

Island says the controls are built natively into the Island Enterprise Browser for Agents and cannot be disabled or bypassed by end users or agents. Governed access to browser tools is the first layer: an agent receives only authorized tools and pages, set per agent and per application, down to which document object model operations are allowed and whether the agent can write or only read. The Model Context Protocol layer is policy-governed as well, so an agent only receives the capabilities it is authorized to use.

Identity binding is the second layer. Agents inherit the enterprise identity of the user and operate under that user's permissions. Policy can apply to the user alone, to the user and their agents together, or to the agents alone, giving administrators a finer dial than a single shared service account. That distinction is becoming a recurring theme in the agent economy, where identity and accountability are now treated as infrastructure rather than an afterthought. Recent coverage of agent identity platforms handling millions of daily calls shows how quickly enterprises are moving to treat agents like employees with verifiable credentials.

Attribution is the third layer. Audit logs record whether a human or an agent took each action, which the company says supports forensic investigations and compliance workflows covering SOC 2, ISO 27001, HIPAA-eligible processes, PCI DSS, and the EU AI Act. Sessions can also be recorded in full for visual playback of every page, click, and form submission, giving reviewers a way to reconstruct exactly what an agent did.

Read-only mode on sensitive systems adds a further guardrail. Agents can work freely in low-risk tools while critical systems stay closed to write privileges. In enterprise resource planning systems and other systems of record, agents view and extract data without modifying, submitting, or deleting it. Data boundaries and credential isolation complete the picture: Island inspects rendered content, clipboard operations, file uploads, print jobs, and screen captures before encryption, and holds agents inside a defined perimeter of sanctioned applications. A built-in vault injects credentials directly into login forms, including applications with no identity provider integration, so passwords never pass through the model context window.

Internal application access without a virtual private network is another feature. Zero Trust Network Access connectors expose only the specific internal applications that policy allows, each gated by identity and device posture, with no lateral network reach. Digital experience monitoring for agents rounds out the package, measuring page load times, interaction latency, errors, and completion rates for agent sessions so teams can find bottlenecks and measure automation return on investment.

Faster agents and lower token bills

Agentic browsing can also run up costs beyond planned budgets, according to Island. Consumer browser extensions that depend on screenshots often require the model to interpret pixels rather than structured page data. With web pages carrying HTML, CSS, and JavaScript, pushing all of that content through an AI model burns tokens at each step.

Island says it owns the browser engine, and its annotation-based application programming interfaces enable efficient page parsing, reducing token consumption and improving interaction reliability. The company also says the browser learns from repeated agent workflows, remembering navigation paths and pre-loading resources so the same multi-step process may complete faster on subsequent runs. For teams running agents at scale, that combination of lower token use and faster repeat runs could matter as much as the security controls.

What the launch signals for the agent economy

Island chief technology officer and co-founder Dan Amiga said the company already had what agents need, including identity federation, data boundaries, credential isolation, and full attribution, all proven in production. In his words, agents inherit the security model of Island on day one, with policies built specifically for them. He also predicted that within a few years, most enterprise browser activity will come from agents, not people, and warned that organizations treating agent browsing as a separate problem end up running a second workforce nobody can see.

That prediction fits a broader shift in 2026, as standards bodies and vendors race to define how agents identify themselves, delegate work, and answer for what they do. The work of the Agentic AI Foundation on agent identity and delegation standards reflects the same concern from the open-standards side: without shared rules for identity and accountability, agent deployments will fragment across incompatible controls. Island is betting that the browser, where so much enterprise work actually happens, is the right place to enforce those rules.

Questions remain, as with any new control plane. Enterprises will want to see how the browser handles complex single-page applications, how policy conflicts are resolved when a user and their agent have different permissions, and how much operational overhead the monitoring and recording features add. Pricing and availability details were not included in the launch announcement, and adoption will depend on how easily the product fits existing identity and security stacks.

Even with those open questions, the launch marks a clear moment: the browser is no longer treated as a neutral window onto the web, but as a governed workspace where agents do their jobs. As agents take on more of the clicking, filing, and form-filling that once belonged to people, the controls around that activity may prove as important as the models driving it.

Sources: reporting on the Island launch from GlobeNewswire and related coverage of agent verification launches on October 29, 2026.