SentinelOne has unveiled Singularity Agentic Detection and Response (ADR), a product that finds AI agents inside an organisation, sets limits on what they can do, and detects and stops unintended or malicious activity as it happens. It is available in early access.

The company introduced the product at OneCon 2026, its global customer conference. It builds on the Singularity Platform and adds to the AI security portfolio, which in March gained Prompt AI Agent Security for discovering agents and governing how they connect to tools and data.

SentinelOne’s argument is that the risk from agents now comes as much from approved agents going off course as from attackers. “Securing AI agents goes beyond just stopping threat actors with malicious intent,” said Chris Corde, Chief Product Officer. “As we’ve seen all too frequently, an equal if not more prominent risk comes from unintended behavior of sanctioned agents.”

According to the company, Singularity ADR offers:

• Inventory: every agent, tool, skill, connector and MCP server across endpoints, SaaS and cloud, with the permissions each holds and the employees behind it.
• Platform coverage: Salesforce Agentforce, Microsoft Copilot Studio, OpenAI ChatGPT workspace agents and Anthropic Claude managed agents.
• Governance and prevention: allow and block controls, approval workflows for high-risk tools, and blocking of prompt injection, jailbreaks, credential exfiltration and tool poisoning before an action runs.
• Detection and response: flagging of “intent drift” between what an agent was asked to do and what it did, a single execution graph linking user request, AI assistant, MCP server and the process that followed, and containment of one agent mid-action without halting the rest of a workflow.
• Investigation: automated investigation and response within limits set by the customer.

“Agents hold real credentials and act without waiting for anyone,” said Tomer Weingarten, co-founder and CEO of SentinelOne, who said the product brings existing runtime protection to “the agents doing real work inside the enterprise.”

The release gives no pricing or general availability date, and no test results or customer deployments for ADR itself. It refers to “recent high profile events” involving agents without naming them. Claims such as stopping rogue agents “at machine speed” are SentinelOne’s and were not independently verified.