Prompt Security, now part of SentinelOne, protects organisations as they adopt AI. It finds the AI tools, applications and agents in use across a business and applies policy to them, so employees and developers can use AI without leaking sensitive data, and homegrown AI apps and agents cannot be turned against the company.
It is for security teams caught between business pressure to adopt AI quickly and the need to know where company data is going and what AI agents are allowed to do.
At A Glance
| Vendor | Prompt Security (from SentinelOne) |
|---|---|
| Category | AI security: AI usage control, agentic AI security, AI application security |
| Identities covered | Employees using AI tools, AI code assistants, AI agents and MCP servers |
| Deployment | Part of the SentinelOne Singularity Platform |
| Agentic AI | Agent and MCP discovery, least-privilege access, audit of agent actions |
| Customers | 10x Banking, Carnival Corporation |
| Market classification | AI Safety & Guardrails; also Agentic AI Identity |
What Prompt Does
Enterprise AI adoption brings a set of risks the platform is designed to contain: shadow AI, sensitive data leakage, prompt injection, jailbreaks, unsafe outputs, data poisoning and unauthorised agent actions. The approach has two steps. First it discovers which AI tools, apps and agents are actually in use, including ones nobody approved. Then it enforces policy on them, from role-based rules on who may use which tool to real-time redaction of data and blocking of hostile prompts. The same approach covers people using AI, agents acting on their behalf and the applications a company builds itself.
Products
AI Usage Control
AI Usage Control is for security teams who need to let staff use AI safely. It builds an inventory of the AI tools and code assistants employees use, including unsanctioned shadow AI, and enforces policy across more than 15,000 AI services in real time, redacting sensitive data before it leaves. Developers keep their coding assistants without exposing IP or secrets. Guardrails include Customer Guardrails, Topics Detector, Sensitive Data and Secrets.
Agentic AI Security
Agentic AI Security gives visibility and control over the agents and MCP servers running in an organisation. It maps agents and MCP servers, uncovers shadow MCP servers and unsanctioned agent deployments, enforces least-privilege access and keeps a searchable audit log of what agents do.
AI Application Security
AI Application Security protects the AI applications a company builds for its own customers or staff. Before deployment it tests them for prompt injection, jailbreaks and data poisoning; in production a real-time AI firewall blocks adversarial prompts and scrubs sensitive outputs. Findings from red teaming become production guardrails, so the same weaknesses stay closed.
Prompt Fuzzer
Prompt Fuzzer is an open-source tool built for AI applications that tests and hardens an app’s system prompt. It uses evolutionary fuzzing, with each iteration run as a new fuzzing campaign. The code is on GitHub as ps-fuzz.
ClawSec And OneClaw
ClawSec is an open-source security suite for OpenClaw, NanoClaw and Hermes agents, with drift detection, live recommendations, automated audits and skill integrity verification. OneClaw is a service that discovers personal AI assistants in an organisation and shows agent activity, usage patterns and security insights in real time.
How It Works
Prompt Security deploys without complex integrations or infrastructure changes, and once running it discovers new AI tools, apps and agents automatically. Every agent action it sees is written to a searchable audit log. Visibility of employee AI use comes from sensors in the browser and on the endpoint, where most AI interactions start. For homegrown applications, the AI firewall sits inline between the application and its users in production. Within SentinelOne, the platform is delivered as part of the Singularity Platform, next to its endpoint, cloud, identity and data security components.
Identities And Systems Covered
| Employees | AI usage under role-based policies. |
|---|---|
| Developers | AI code assistants such as GitHub Copilot, Cursor and Claude Code. |
| AI agents | Mapping, shadow agent discovery, least-privilege access and audit. |
| MCP servers | Mapping and shadow MCP discovery, with MCP gateway security between AI applications and more than 13,000 known MCP servers. |
| Secrets and sensitive data | Redaction and a dedicated Secrets guardrail. |
Company Background
| Leadership | Itamar Golan (CEO and co-founder). |
|---|---|
| Acquisition | SentinelOne announced on 5 August 2025 that it would acquire Prompt Security for a combination of cash and stock. |
| Open source | Tools are published under the prompt-security GitHub organisation. |
Note: This overview is compiled from Prompt’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.