SentinelOne announced on October 21 at its OneCon 2026 conference that Singularity Agentic Detection and Response (ADR) is available in early access, extending the company’s runtime behavioral security to AI agents running across endpoints, cloud workloads, SaaS platforms and identity systems. Help Net Security reported that the offering pairs visibility and governance of agent activity with SentinelOne’s existing prevention, detection and response stack, built directly on the Singularity platform.
The launch is a product announcement rather than a vulnerability disclosure or incident response: SentinelOne cited no specific attacks, victims or threat actors. The company said AI agents increasingly hold credentials and act independently, while attackers use frontier models to find and chain vulnerabilities, and framed ADR as defense “at machine speed” for the devices where AI work runs.
What ADR does
According to SentinelOne’s press release, the product maintains an inventory of every agent, tool, skill, connector and Model Context Protocol (MCP) server across endpoints, SaaS and cloud providers, including each one’s permissions and the employees behind them. Coverage extends beyond MCP gateways to Salesforce Agentforce, Microsoft Copilot Studio, OpenAI ChatGPT workspace agents and Anthropic Claude managed agents.
SentinelOne said ADR monitors every agentic event and sends it to the Singularity Data Lake as “Agent Traces,” where it can be searched, visualized and used in detection rules. A “Unified Storyline” view folds agent events into the same execution graph as traditional endpoint detection and response, tracing the chain from user request to AI assistant to MCP server to the resulting file or process.
On the enforcement side, ADR’s in-line prevention blocks prompt injection, jailbreaks, credential exfiltration and tool poisoning before an action executes. Policy controls allow or block which agents can run, who can use them and what they can reach, with approval workflows for high-risk tools. An “intent drift detection” feature flags gaps between what an agent was asked to do and what it actually did, and agent containment can stop a single agent mid-action while the rest of a workflow continues.
Broader AI security push and open questions
ADR is the newest piece of a wider portfolio. SentinelOne positioned it as a step beyond the agent discovery and governance capabilities the company introduced in March 2026. In parallel, it unveiled Singularity ASOC, an integrated suite powered by its “Purple Security Superintelligence” that combines real-time data ingestion, agentic investigation and automated response, with analysts keeping control over key decisions under what the vendor calls a “human in command” model.
A CRN report from OneCon cited SentinelOne CEO Tomer Weingarten saying more than 5,000 customers have used the company’s agentic investigation capability, cutting investigation times from roughly 50 minutes to about five. That figure describes the existing Purple capability, not ADR, which has no customer metrics yet. SentinelOne also announced an AI Security partner program, with certified partners able to join the ADR early access program and an AI Recon Assessment Service for customer engagements.
SentinelOne faces competition in this space: Palo Alto Networks markets Cortex XSIAM as an AI-driven SOC platform with its own “Agentic SOC” concept and claims of cutting manual work by up to 75 percent.
What to do
- Organizations deploying AI agents can enroll in the ADR early access program through SentinelOne or a certified partner; general availability timing and pricing were not disclosed.
- Start with the agent and MCP inventory to find unsanctioned agents. A SentinelOne technical blog describes using Singularity and Data Lake PowerQuery to detect shadow AI assistants such as OpenClaw, Moltbot and Clawdbot through process execution, file activity, network activity and persistence mechanisms, and its open-source ClawSec suite for hardening the OpenClaw agent runtime.
- Apply MCP server hardening in any environment ADR monitors. The Model Context Protocol security best practices warn of “confused deputy” attacks via MCP proxy servers and recommend per-client consent storage and strict token validation.
- SentinelOne’s Prompt Security offering inventories more than 15,000 AI services and enforces per-user, per-team and per-tool policy with real-time redaction of sensitive data, according to the company’s product page.
SentinelOne’s claims are as of October 22 and are largely unverified: no independent evaluations, red-team results or production customer case studies for ADR or ASOC have been published, and the company has not detailed how intent drift detection works, what prompt and tool-call content is retained in the Data Lake or for how long, or how ADR interoperates with non-SentinelOne tooling in multi-cloud environments.