Darktrace has announced the general availability of Darktrace / SECURE AI, a platform designed to help enterprises identify how their employees and agents are using artificial intelligence, detect unauthorized services, and enforce security policies around prompts, data, and AI systems. The company says more than 80% of its customers already use generative AI services, and that in August 2026 the average organization interacted with five different AI providers.
Darktrace / SECURE AI in 30 seconds
- SECURE AI lets organizations discover and control enterprise use of artificial intelligence services.
- Darktrace adds prompt analysis, policy management, and AI agent oversight.
- The platform integrates with AWS, Anthropic, Microsoft, and OpenAI, among other providers.
- Darktrace says more than 80% of its customers use generative AI.
- The solution is available as a standalone product, within its security platform, or through AWS and Microsoft Marketplace.
The launch responds to a scenario in which enterprise AI adoption is advancing in a distributed way. Employees can use different assistants and platforms, while security teams need to know which services are in use, what information is being fed into them, and which agents have access to corporate resources — a challenge Akamai has also targeted with its own Workforce Protector tool.
Darktrace calls the use of AI tools that haven’t been formally authorized or overseen by the organization Shadow AI. SECURE AI aims to provide visibility into this activity and tie it back to each company’s internal policies.
The company says 80% of its monitored customers used some form of generative AI service during August 2026. The figure comes from aggregated telemetry across roughly 8,200 deployments observed by Darktrace, so it reflects the company’s monitored customer base rather than a measurement of the entire enterprise market.
Control over prompts, agents, and AI services
Darktrace / SECURE AI brings together several functions aimed at overseeing how artificial intelligence is used within an organization.
One is Shadow AI management. The system identifies activity tied to unauthorized AI services and can use Darktrace telemetry along with SASE integrations, including Microsoft Entra Global Secure Access, to help block certain uses or quarantine affected devices.
The platform also adds prompt analysis. This feature provides visibility into instructions, sessions, and responses across various supported platforms, including Amazon Bedrock, ChatGPT Enterprise, Claude, Microsoft Copilot, and Copilot Studio. The company says it will add support for Salesforce soon.
The analysis is designed to detect behaviors such as jailbreak attempts, exposure of sensitive information, and possible indirect prompt injection attacks. Sessions are ranked by risk level to help security teams decide which ones need priority investigation.
The solution also includes a policy management feature that lets teams upload documents containing each organization’s specific rules. Administrators can use them to define controls and check their alignment with certain regulatory and compliance frameworks.
This approach carries internal policies over into the day-to-day use of AI systems. Rather than simply identifying which tools are in use, the goal is also to check whether user and agent activity matches authorized uses.
AI agents add another layer of oversight
The growth of AI agents introduces an additional challenge, one that Fortinet has also addressed by extending FortiEndpoint with AI agent identification and inspection. These systems can have their own identities and access applications, data, or infrastructure to carry out tasks.
SECURE AI identifies agents and users tied to AI activity across supported environments and provides information on permissions, roles, access, and relationships. The goal is to help determine what each agent is doing and whether its actions match the purpose it was built for.
Darktrace also adds specific capabilities for monitoring agent development. The platform can analyze low-code and high-code development environments, including Amazon Bedrock and Microsoft Copilot Studio, to identify agent identities, excessive permissions, misconfigurations, and anomalous behavior during creation.
This information connects two different moments in an agent’s lifecycle: how it’s developed and how it behaves once deployed. For security teams, that link makes it possible to check whether an initial configuration ends up producing behavior that wasn’t anticipated.
The company bases this approach on its Adaptive AI technology, which, according to Darktrace, learns each organization’s normal operating patterns. SECURE AI extends that approach to AI users, prompts, agents, and systems.
Darktrace argues that this model can detect behavior that doesn’t necessarily match previously known threats. That claim is part of the company’s technology pitch and doesn’t mean every incident can be detected automatically.
Integrations with AWS, Anthropic, Microsoft, and OpenAI
The launch includes integrations with several of the leading AI infrastructure and model providers.
With Amazon Web Services (AWS), Darktrace expands oversight of the development and operation of agents built with Amazon Bedrock. The idea is to connect how these agents are built with how they behave once deployed.
On Anthropic, SECURE AI adds coverage for Claude’s prompts and responses. The platform lets teams analyze that activity within the context of each organization’s policies and intended use.
The Microsoft integration expands visibility into Microsoft Copilot users and agents, including agent development through Copilot Studio.
With OpenAI, Darktrace offers oversight of ChatGPT Enterprise and Codex usage. The two companies are also working through the Daybreak Defense Network on capabilities that combine OpenAI’s Daybreak models with Darktrace’s behavioral analysis.
According to information published by Darktrace, this work aims to identify incidents related to enterprise AI and add context about the people, systems, and business processes involved. The goal is to help defenders prioritize their responses.
The company also shares several examples from its early customers. At one organization that had authorized a single AI assistant, SECURE AI found that most employees were using unauthorized services instead. At another, it detected several AI platforms being used by contractors with no centralized management, which led to a legal review and the rollout of formal controls.
In a third case, the platform identified close to 90 AI agents within a low-code development environment that had no formal approval or oversight process for their creation.
These examples come from deployments described by Darktrace and illustrate the kinds of situations the company aims to detect. They don’t represent an independent measurement of how often these issues occur across enterprises.
SECURE AI has been available since September 22, 2026, for both new and existing Darktrace customers. It can be purchased as a standalone product or as part of the Darktrace Behavioral Defense Platform, and it’s also available through AWS Marketplace and Microsoft Marketplace.
The company, which says it protects close to 10,000 customers worldwide, positions SECURE AI as a new layer for overseeing the growth of enterprise AI. The focus shifts from controlling only the tools in use to also observing the behavior of the users, agents, prompts, and systems connected to them.
FAQ
What is Darktrace / SECURE AI?
It’s a Darktrace solution for discovering and overseeing enterprise use of artificial intelligence, including authorized and unauthorized services, prompts, users, and agents.
What is Shadow AI?
Shadow AI refers to the use of artificial intelligence tools and services within an organization without formal authorization or oversight from the teams responsible for them.
Which AI platforms does SECURE AI integrate with?
Darktrace lists integrations with Amazon Web Services, Anthropic, Microsoft, and OpenAI, along with support for various enterprise platforms such as Amazon Bedrock, Claude, ChatGPT Enterprise, Microsoft Copilot, and Copilot Studio.
Is Darktrace / SECURE AI available now?
Yes. Darktrace announced its general availability on September 22, 2026, for new and existing customers, as a standalone product, integrated into its security platform, and through AWS Marketplace and Microsoft Marketplace.
