SentinelOne is attempting to solve the emerging security gap created by autonomous AI agents that possess real credentials and the ability to execute independent tool calls across enterprise environments. By launching Singularity Agentic Detection and Response (ADR), the company is moving beyond simple discovery and governance into the realm of real-time runtime protection. This strategic shift addresses a critical vulnerability: the risk of sanctioned AI agents "going off course" or behaving unintentionally. The new offering, unveiled at OneCon 2026, aims to provide security teams with the ability to monitor, govern, and stop individual agentic actions at machine speed without disrupting broader business workflows.
Expanding AI Security from Governance to Runtime
The introduction of Singularity ADR represents a significant expansion of SentinelOne’s AI security portfolio, building upon the Prompt AI Agent Security capabilities released in March. While previous tools focused on discovering AI agents and governing their connections to data, Singularity ADR integrates behavioral detection and response directly into the agentic lifecycle. The company is positioning this as a way to bridge the gap between knowing an agent exists and controlling what that agent actually does once it begins interacting with endpoints, cloud workloads, or SaaS environments.
SentinelOne is targeting a landscape where employees act as "agent managers" using tools like Claude Code, Cowork, and Codex. Because these agents can chain tool calls and operate independently, the company argues that the risk of unintended behavior from a sanctioned agent is as prominent as a direct attack from a malicious actor. To counter this, Singularity ADR utilizes the Singularity Platform to tie every agent action back to its origin. This allows security analysts to view a unified execution graph that connects a user request to the AI assistant, the Model Context Protocol (MCP) server, and the resulting file or process. By integrating agentic events with traditional Endpoint Detection and Response (EDR) data, the company aims to eliminate the need for manual event reconstruction during an investigation.
Technical Capabilities of Singularity ADR
Singularity ADR provides a multi-layered technical approach to securing the "agentic" enterprise. A core component is the inventory of agents, tools, skills, connectors, and MCP servers across endpoints and cloud providers. This includes both sanctioned and unsanctioned agents, providing visibility into the permissions each holds and the specific employees utilizing them. The platform's coverage extends to major ecosystems, including Salesforce Agentforce, Microsoft Copilot Studio, OpenAI ChatGPT workspace agents, and Anthropic Claude managed agents, ensuring that security follows the agent regardless of where the business deploys it.
For active defense, the offering implements in-line prevention to block prompt injection, jailbreaks, credential exfiltration, and tool poisoning before an action can execute. A key technical distinction is the "intent drift detection" feature, which flags discrepancies between an agent's original instructions and its actual system behavior. If an agent deviates from its intended path, security teams can utilize agent containment to stop a single agent mid-action. This granular control is designed to prevent a rogue process from spreading while allowing the rest of the enterprise workflow to continue uninterrupted. All agentic events are routed to the Singularity Data Lake for visualization, searching, and long-term retention within security dashboards.
Key Takeaways
- SentinelOne has launched Singularity Agentic Detection and Response (ADR), which is currently available in early access.
- The platform provides multi-platform coverage for agents running in Salesforce Agentforce, Microsoft Copilot Studio, OpenAI ChatGPT, and Anthropic Claude.
- Singularity ADR enables "agent containment," allowing security teams to stop a single autonomous agent mid-action without halting the entire workflow.
TechInsyte's Take
In our view, SentinelOne is making a calculated bet that the next major frontier in cybersecurity is not just the models themselves, but the autonomous "hands" that models use to interact with enterprise data. As organizations move from passive chatbots to active AI agents capable of executing code and managing credentials, the attack surface shifts from data leakage to unauthorized autonomous action. By linking agentic behavior to their existing EDR and cloud security telemetry, SentinelOne is attempting to prevent the "siloing" of AI security. This approach suggests that the company views AI agent security not as a standalone niche, but as a necessary evolution of endpoint and identity protection. If successful, this integration could set a standard for how enterprises manage the inherent unpredictability of autonomous software agents in production environments.
Questions & Answers
How does Singularity ADR differentiate itself from previous AI security offerings?
While previous offerings like Prompt AI Agent Security focused on the discovery and governance of AI agents, Singularity ADR adds a runtime layer. It moves beyond just knowing which agents are connected to which data to actively monitoring and stopping agentic behavior in real time using behavioral detection.
What specific types of AI agent environments does the platform cover?
The platform provides coverage across multiple environments, including Salesforce Agentforce, Microsoft Copilot Studio, OpenAI ChatGPT workspace agents, and Anthropic Claude managed agents, as well as various MCP (Model Context Protocol) servers.
Can security teams stop a single rogue agent without impacting the entire business process?
Yes. One of the specific capabilities of Singularity ADR is agent containment, which allows security teams to stop a single agent mid-action while the rest of the workflow continues to run.
How does the platform help analysts investigate an AI-driven security event?
The platform uses a "Unified Storyline" to integrate agentic behavior with traditional EDR data into a single execution graph. This allows analysts to see the entire chain of events—from the initial user request and the AI assistant to the specific tool or process that followed—in one place.
Source: SentinelOne
