Darktrace / SECURE AI
by Darktrace
Behavioral monitoring of enterprise AI use: shadow AI blocking, prompt risk analysis and AI agent identity mapping
Darktrace / SECURE AI is a security product for enterprises that want to see and control how employees and AI agents use generative AI. It detects shadow AI, analyses prompts and sessions on ChatGPT Enterprise, Claude, Microsoft Copilot and Amazon Bedrock for jailbreaks and data exposure, and maps agents to the people and permissions behind them.
Darktrace / SECURE AI applies Darktrace's behavioral detection, which learns what is normal for each organization, to enterprise use of generative AI and AI agents. Darktrace announced it in February 2026 and made it generally available on September 22, 2026, sold standalone or as part of the Darktrace Behavioral Defense Platform and listed on AWS Marketplace and Microsoft Marketplace. It has five components. Shadow AI Management finds unsanctioned AI services from Darktrace telemetry and SASE integrations such as Microsoft Entra Global Secure Access, and can block the service or quarantine the device. AI Prompt Analysis captures prompts, sessions and responses on Amazon Bedrock, ChatGPT Enterprise, Claude, Microsoft Copilot and Copilot Studio, flags jailbreak attempts, sensitive data exposure and possible indirect prompt injection, and ranks sessions by risk; Salesforce support was listed as coming soon at launch. Policy Management lets administrators upload free-form AI policies and check them against regulatory frameworks. AI Agent Identities and Actions links agents to the human users behind them and their permissions, roles and connections, including MCP connections to external services. AI Agent Development Risk Management watches low-code and high-code build environments such as Bedrock and Copilot Studio for excessive permissions and misconfigurations. Darktrace says telemetry from about 8,200 deployments showed more than 80% of customer accounts using generative AI services in August 2026, with the average organization using five AI providers. Darktrace has been privately owned by Thoma Bravo since an all-cash acquisition of about $5.3 billion closed on October 1, 2024, and says it protects nearly 10,000 customers.
A security operations team, ideally one already running Darktrace, that needs prompt-level visibility and response for sanctioned AI assistants and the agents being built on Bedrock and Copilot Studio.
One behavioral view of shadow AI, risky prompts and agent permissions, with the option to block a service or quarantine a device from the same console.
At a Glance
- Category
- Governance & Security
- Pricing
- Subscription, Contact for pricing
- Target Market
- CISOs, Security Operations Teams, CIOs, AI Governance Teams
- Deployment
- Cloud-first, API-based
- Founded
- 2013
- Headquarters
- Cambridge, United Kingdom
- Team Size
- 500+
- Customers
- Nearly 10,000 customers across the Darktrace platform (company figure, September 2026)
Key Features
- ✓Shadow AI Management
Detects unsanctioned AI services from Darktrace telemetry and SASE integrations, then blocks the service or quarantines the affected device.
- ✓AI Prompt Analysis
Captures prompts, sessions and responses on supported platforms, flags jailbreaks, data exposure and indirect prompt injection, and ranks sessions by risk.
- ✓Policy Management
Lets administrators upload free-form AI policies, define custom controls and review them against regulatory and compliance frameworks.
- ✓AI Agent Identities and Actions
Maps each AI agent to the human users behind it plus its permissions, roles and MCP connections to external services.
- ✓AI Agent Development Risk Management
Watches agent creation in Bedrock and Copilot Studio for excessive permissions and misconfigurations before those agents reach production.
- ✓Darktrace MCP server
Connects compatible AI assistants to Darktrace through the Platform Portal, with a read-only mode that hides tools able to change data.
Use Cases
- •Copilot and ChatGPT Enterprise prompt monitoring
The SOC reviews risk-ranked sessions where employees pasted sensitive data or attempted jailbreaks, and investigates the highest-risk users first.
- •Low-code agent sprawl
Security finds dozens of unapproved agents built in a low-code environment, reviews their permissions and removes the excessive ones.
- •Contractor shadow AI
Unmanaged AI use by contractors is detected through SASE telemetry, the service is blocked and the finding goes to legal review.
- •Policy-to-practice gap analysis
Governance uploads the company's AI policy, compares it against observed prompt activity and frameworks, and closes the gaps it finds.
Ideal For
Best For
- ✓Existing Darktrace customers extending network and email behavioral detection to AI assistants and agents
- ✓Prompt and session monitoring for ChatGPT Enterprise, Claude, Microsoft Copilot and Amazon Bedrock for jailbreaks and data exposure
- ✓Finding over-permissioned or misconfigured agents as they are built in Copilot Studio or Bedrock
- ✓Blocking unsanctioned AI services through Microsoft Entra Global Secure Access and other SASE integrations
- ✓Checking written AI use policies against regulatory and compliance frameworks
Not Ideal For
- ✗Organizations whose AI use sits outside the supported platforms or runs on local models; prompt-level coverage is limited to the named integrations
- ✗Budget-constrained teams: PeerSpot reviewers of the wider Darktrace platform describe annual costs from about $30,000 to over $100,000
- ✗Teams without analyst capacity to tune detections, since Darktrace reviewers repeatedly report false positives on its anomaly models
Deployment
Market & Ratings
Nearly 10,000 customers across the Darktrace platform (company figure, September 2026)
Market Analysis
Pros
- ✓Prompt-level visibility across the main enterprise assistants and Bedrock from a single console
- ✓Can act as well as alert: block unsanctioned services or quarantine devices through platform integrations
- ✓Large installed base; PeerSpot rates the wider Darktrace platform 4.1/5 across 85 reviews, with 95% willing to recommend
Cons
- ✗False positives are the most common technical complaint in PeerSpot reviews of Darktrace's anomaly detection, and no detection-accuracy figures were published for SECURE AI
- ✗Expensive: PeerSpot reviewers cite $30,000 to over $100,000 a year for the platform, and one estimate of about $350,000
- ✗Coverage is limited to supported platforms; an independent analysis notes unclear visibility for encrypted traffic and local-model use and advises starting in observe-only mode
- ✗Reviewers report integration friction with some firewalls, SIEMs and email gateways
Pricing
Darktrace / SECURE AI
Contact for pricing
- ✓Standalone or as part of the Darktrace Behavioral Defense Platform
- ✓Available on AWS Marketplace and Microsoft Marketplace
Darktrace does not publish a price for SECURE AI; it is quoted through sales or bought through AWS or Microsoft Marketplace. For context, PeerSpot reviewers of the broader Darktrace platform report annual costs from about $30,000 to over $100,000, licensed by the number of devices ingesting data, on annual or three-year terms, with discounts available through negotiation.
Security & Compliance
Sources
This page was written from 7 sources, 5 on domains other than darktrace.com.
- 1.darktrace.com — darktrace extends behavioral security with general availabilvendor
- 2.darktrace.com — darktrace secure ai extending behavioral security for the agvendor
- 3.siliconangle.com — siliconangle.com
- 4.securitymea.com — securitymea.com
- 5.lapaasvoice.com — darktrace secure ai general availability
- 6.peerspot.com — darktrace reviews
- 7.thomabravo.com — thoma bravo completes acquisition of darktrace
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Abnormal AI Governance
Shadow AI discovery, AI agent inventory and AI spend tracking built from email, identity and OAuth signals
Akeyless Agentic Runtime Authority
Intent-based runtime access control for AI agents: credentialless access, per-action policy checks and a session kill switch
Pillar Security
AI security platform for discovering, red-teaming and guarding AI agents, MCP servers and coding agents, being acquired by Vanta
Bay
AI agent security platform that discovers every agent and MCP server and blocks risky agent actions at runtime