Prompt Security Review (2026): SentinelOne's AI usage control, MCP gateway and app protection
Prompt Security is a Tel Aviv AI security company that SentinelOne bought in September 2025. It watches how employees, developers, homegrown apps and agents use AI, redacts sensitive data before it leaves the device, and puts an MCP gateway in front of agent tool calls. It is one of the most complete products for workforce AI control, and it fits best if you already run SentinelOne.
- Headquarters
- Tel Aviv, Israel
- Founded
- 2023
- Owner
- Part of SentinelOne (NYSE: S) since September 2025
- Fits best
- Security teams, especially SentinelOne customers, that need to see and control employee, developer and agent use of AI tools with data redaction and MCP policy
- Products
- Prompt Security for Employees (AI Usage Control), Prompt Security for AI Code Assistants, Prompt Security for Homegrown Apps (AI Application Security), Prompt AI Agent Security with MCP Gateway, Prompt AI Red Teaming
- Pricing
- Quote only. The AWS Marketplace listing shows six 12-month options at a $10,000 list figure each and points buyers to private offers.
Overview
What Prompt Security does
Prompt Security was started in Tel Aviv in August 2023 by Itamar Golan (CEO) and Lior Drihem (CTO). It raised an $18 million Series A in November 2024, led by Jump Capital, for about $23 million in total funding. SentinelOne signed a deal to buy it on August 5, 2025 and closed on September 5, 2025. SentinelOne's quarterly filing puts the fair value of the consideration at about $159.3 million, mostly cash plus 1,555,099 Class A shares.
The product began as a way to see what employees paste into ChatGPT and similar tools. It now covers four areas: workforce AI usage (SentinelOne calls this AI Usage Control), AI code assistants such as GitHub Copilot, Cursor and Claude Code, the AI apps a company builds itself, and AI agents with their MCP servers. Collection happens through a browser extension, a lightweight endpoint agent, an MCP gateway and API integrations, and SentinelOne says it recognizes more than 15,000 AI tools and services.
Since the deal, the Prompt name has stayed as a product brand inside the Singularity Platform. Agent security runs in the same console SentinelOne uses for endpoints, identity and cloud, which is the main reason SentinelOne shops shortlist it. Buyers on other endpoint vendors can still buy it on its own, including a self-hosted edition.
Strengths
- Covers employee AI use, code assistants, homegrown apps and agents in one product
- Inline redaction and coaching let people keep using AI instead of being blocked outright
- Dedicated MCP gateway with risk scoring and an audit log of agent actions
- Self-hosted option for each core module, plus SentinelOne console and endpoint context
Watch out for
- Strongest fit is inside SentinelOne; buyers on other endpoint stacks lose part of the value
- Coverage numbers such as 15,000 AI tools and MCP servers come from the vendor
- Several separate SKUs, so total cost depends on how modules and user counts are bundled
Features
Prompt Security feature by feature
Browser extension and shadow AI discovery
The browser extension and endpoint agent detect AI use in browsers, desktop apps, IDEs and terminal assistants, sanctioned or not, and keep a searchable log of prompts and responses. Security teams see which tools each user and department uses. Mobile devices are outside its scope; SentinelOne points to Singularity Mobile for those.
Inline redaction and user coaching
Policies can redact or tokenize personal data, secrets and other sensitive content before a prompt leaves the device, block high-risk prompts, or show the user a warning and let them carry on. That middle option matters for adoption, because blanket blocking tends to push people toward personal accounts.
MCP gateway for agents
Prompt Security's MCP gateway sits between AI applications and MCP servers and inspects each call, prompt template and response. SentinelOne says it risk-scores more than 15,000 MCP servers and can allow, block, filter or redact. Agent and MCP server discovery covers unsanctioned servers, with a relationship map and an audit trail of what each agent touched.
AI code assistant controls
A dedicated module watches Copilot, Cursor and Claude Code sessions to stop secrets, credentials and proprietary code from leaking into prompts. It is sold as a separate SKU, so check whether your developer count or your employee count drives the bill.
Homegrown app firewall and red teaming
For apps you build, a runtime firewall checks prompts and responses for prompt injection, jailbreaks, data leakage and off-brand output, which the vendor says deploys with a single line of code. Prompt AI Red Teaming, announced at RSAC 2026, tests apps before release and feeds findings into the runtime rules.
Deployment options
The platform is sold as SaaS or self-hosted for each of the three main modules, which helps teams that cannot send prompt content to a vendor cloud. It also has open source projects, including the ps-fuzz prompt fuzzer.
Pricing
Prompt Security pricing
SentinelOne does not publish a Prompt Security price list. The AWS Marketplace listing, sold by SentinelOne, shows six 12-month dimensions (Employees, AI Code Assistants and Homegrown Apps, each as SaaS or self-hosted) at a $10,000 list figure each, and every line tells buyers to contact SentinelOne for custom pricing and private offers. Treat that number as a placeholder, since the listing does not say what one unit covers.
In practice the bill scales with covered users per module: employees for workforce AI control, developers for code assistant protection, and app or traffic scope for homegrown apps. Ask whether agent security and red teaming are separate SKUs, whether existing Singularity customers get bundle pricing, and what the self-hosted edition adds in support cost.
Alternatives
Prompt Security alternatives
When another vendor may fit better.
Pick it if CrowdStrike Falcon is your endpoint platform and you want AI detection and response from the former Pangea team in the same console.
Pick it if you mostly need runtime guardrails and red teaming for AI apps you host, including on premises or air-gapped.
Pick it if Zscaler already inspects your web traffic and you would rather enforce AI usage rules at the proxy than with an extension.
Pick it if you want shadow AI discovery and DLP folded into an existing SSE or CASB deployment.
Pick it if your main worry is sensitive data in employee prompts and you want a lighter, data-focused tool.
Pick it if your agents live mostly in Copilot Studio, Salesforce and other low-code platforms and need posture plus runtime control there.
Verdict
Should you buy Prompt Security?
Prompt Security is one of the most complete products for controlling how a workforce uses AI, and its MCP gateway puts it ahead of many rivals on agent tool calls. It is an easy pick for SentinelOne customers and a reasonable one for others who need a self-hosted option. Before you sign, test redaction accuracy and browser coverage on your own users, and get the per-module pricing in writing.
What is Prompt Security?
Prompt Security is an AI security platform that discovers and controls how employees, developers, homegrown apps and AI agents use generative AI. It was founded in Tel Aviv in 2023 and has been part of SentinelOne since September 2025.
Is Prompt Security part of SentinelOne?
Yes. SentinelOne announced the acquisition on August 5, 2025 and completed it on September 5, 2025. Its filing values the consideration at about $159.3 million. The product is now sold as Prompt Security from SentinelOne within the Singularity Platform.
How does the Prompt Security browser extension work?
The extension, together with a lightweight endpoint agent, detects AI tools in use, captures prompts and responses with context, and applies policy inline. It can redact sensitive data, block a prompt or coach the user before anything reaches the AI service.
What is the Prompt Security MCP gateway?
It is a control point between AI applications and MCP servers. It inspects each call, prompt template and response, risk-scores servers, applies allow, block, filter or redact rules, and keeps an audit log of agent actions, including on servers nobody approved.
How much does Prompt Security cost?
Pricing is quote based. The AWS Marketplace listing shows six 12-month options at a $10,000 list figure each, covering employees, code assistants and homegrown apps in SaaS or self-hosted form, but buyers are directed to private offers sized by users.
Who are Prompt Security's competitors?
Common alternatives are CrowdStrike Falcon AIDR, F5 AI Guardrails, Zscaler, Netskope, Harmonic Security and Zenity. The choice usually depends on which endpoint, proxy or agent platform you already run.
Free and independent
Comparing Prompt Security with other vendors?
Tell us what you need to protect and we will send three vendors that fit, including whether Prompt Security is one of them.