Island launches enterprise browser to control AI agents

Island has launched an Enterprise Browser for Agents, aimed at organisations that want to apply existing browser controls to AI agents.

The software is designed to let agents access software-as-a-service tools, internal applications and public websites while operating under enterprise policies. Island presents it as a way to extend the controls already used for human employees to software agents that browse, read and act across systems.

Businesses are beginning to use agents for routine work such as updating customer records, onboarding staff and processing invoices. Those tasks often require access to web applications, internal sites and online forms, creating a need for the same kind of oversight that applies when staff use a browser.

Many organisations have so far relied on patchwork arrangements built on consumer browser infrastructure, which Island argues can create risks around credentials, data handling and changes to records.

"Island already had what agents need: identity federation, data boundaries, credential isolation, and full attribution, all proven in production," said Dan Amiga, Chief Technology Officer and Co-founder at Island. "Agents inherit Island's security model on day one, with policies built specifically for them."

Its controls apply at the browser activity level rather than attempting to govern the agents themselves. Under that model, policy is enforced when an endpoint or cloud agent calls the browser or accesses internal resources, including controls over data movement, credential use and outputs.

The product can be deployed on devices and in the cloud, covering both local and remote agents without relying on a cloud-hosted browser running outside the customer's environment.

Built-in controls

A core part of the launch is a set of controls built into the browser engine. Island says these controls cannot be disabled or bypassed by end users or agents.

These include governed access to browser tools and pages, with permissions set by agent and by application. Policy can define which document object model operations are allowed and whether an agent can write or only read. The model context protocol layer is also governed by policy, so agents receive only authorised functions.

The browser also ties agent activity to enterprise identity. According to Island, agents inherit the user's identity and permissions, while policy can be applied to the user alone, to the user and agent together, or to the agent separately.

Audit and attribution are another part of the offering. Logs record whether an action was taken by a human or an agent, and sessions can be recorded for visual playback of pages, clicks and form submissions.

Island is also offering a read-only mode for sensitive systems. This is intended to let agents extract information from systems of record such as enterprise resource planning platforms without being able to modify, submit or delete data.

On access and data handling, the browser can inspect rendered content, clipboard operations, file uploads, print jobs and screen captures before encryption. It also includes a credential vault that injects passwords directly into login forms, including for applications without identity provider integration, so credentials do not pass through the model context window.

For internal application access, Island says its zero-trust network access connectors expose only approved applications and do not allow wider lateral movement across a network. It has also included monitoring tools that measure page load times, interaction latency, errors and completion rates in agent sessions.

Cost focus

The launch also reflects concern about the cost of agentic browsing. Island argues that browser extensions relying on screenshots can force models to interpret pixels rather than structured page data, while sending HTML, CSS and JavaScript through models can increase token use.

Its browser engine uses annotation-based application programming interfaces for page parsing, which Island says reduces token consumption and can improve the reliability of interactions. The company also says the system can learn from repeated workflows by remembering navigation paths and pre-loading resources for recurring tasks.

The announcement comes as technology suppliers race to define how AI agents should operate inside corporate systems. For many employers, the issue is shifting from whether agents can complete tasks to how their actions are identified, limited and recorded across the same applications used by staff.

Island says it has applied the same policy framework used in its enterprise browser for human workers to this new product for agents. It also says it helped define the enterprise browser category.

"Within a few years, most enterprises' browser activity will come from agents, not people," said Amiga. "Organisations that treat agent browsing as a separate problem end up running a second workforce nobody can see. Whether an agent needs a browser, local desktop resources, or upstream MCPs and LLMs, Island is there."