If your team has approved one AI tool, there’s a good chance people are already using several others. Add contractors, low-code builders and a few helpful automations, and suddenly nobody is fully sure which model touched what data.
That’s no longer just an IT housekeeping issue. It’s becoming a day-to-day operating risk for any business that wants AI to help without letting usage sprawl past policy.
Darktrace turns AI use into something security teams can actually see
The core message in SiliconANGLE’s coverage is simple: AI use inside companies is already broader than most leaders think. Darktrace says telemetry from roughly 8,200 deployments showed more than 80% of monitored customer accounts used generative AI services in August, with the average organization interacting with five AI providers that month.
Its new generally available SECURE AI product is built around behavioral monitoring of that activity. The release adds integrations with AWS, Anthropic, Microsoft and OpenAI environments, and it’s on sale now for new and existing customers, including through AWS Marketplace and Microsoft Marketplace.
- General availability. Darktrace made SECURE AI generally available as a standalone product and within its Behavioral Defense Platform, after first announcing it in February.
- Real-time session monitoring. The product scans sessions in ChatGPT Enterprise, Claude, Microsoft Copilot and Amazon Bedrock for jailbreak attempts, sensitive data exposure and signs of indirect prompt injection, then ranks sessions by risk.
- Shadow AI controls. Darktrace says admins can detect unsanctioned AI services through SASE-related integrations, block them, or quarantine a device when needed.
- Agent and build-time visibility. SECURE AI maps agents to the people behind them and extends into development environments such as Amazon Bedrock and Microsoft Copilot Studio to catch excessive permissions and misconfigurations while agents are still being built.
The hard part isn’t one chatbot
For a business owner, the risk isn’t only that employees might paste sensitive information into the wrong tool. It’s that AI usage fragments across approved apps, personal habits, contractor workflows and homegrown agents faster than policy, oversight and training can keep up.
That creates a management problem before it creates a breach headline. You can’t set sensible rules, approvals or access boundaries if you don’t know which assistants are being used, which documents they can reach, or who created the agent now touching internal work.
Darktrace’s launch matters because it treats AI activity as something observable and governable, not just something employees are experimenting with. The stronger signal here is operational: businesses are moving from asking whether people use AI to asking where it’s used, by whom and under what limits.
That shift tends to separate useful adoption from messy adoption. The companies that benefit most will be the ones that narrow the number of places work happens, define clear boundaries and make approved workflows easier than workarounds.
Where Metomorph fits
Metomorph speaks to that same business need from the workflow side: give teams one governed place to do AI-assisted work instead of letting context, documents and prompts scatter across disconnected tools.
Imagine your team is preparing client deliverables with AI help. In Metomorph, that work can live inside a Project, with access limited to the people on that engagement and the relevant files connected only to that workspace.
Project context means the assistant already works from the project’s notes, knowledge and constraints, so staff don’t have to keep re-pasting background into different chats. That reduces the temptation to bounce between multiple outside tools just to recreate context.
You can also set business rules in plain language, such as requiring approval before certain output is sent or restricting how sensitive client information is handled. Those rules apply across chats, skills, apps, reports and forms, which is useful when the real problem is consistency, not just model quality.
A practical next step is to identify one high-value workflow, move it into a single governed workspace, and make that approved path easier than ad hoc AI use. That won’t replace security monitoring, but it can reduce sprawl before it becomes another system to police.
What to do this week
- Count your actual AI touchpoints, not just your approved ones. Include employees, contractors, low-code agents and any connected model providers.
- Pick one or two approved workflows where AI is clearly useful, then give them a defined home, clear access boundaries and written rules for use.
- Review who can create or connect automations. Agent sprawl often starts as convenience, not intent.
- This week, choose one sensitive process and map where its AI-assisted work happens now versus where it should happen under policy.
Explore Projects, Project context, Business rules in Metomorph.
Source: SiliconANGLE, Exclusive: Darktrace makes SECURE AI generally available to monitor enterprise AI use (September 22, 2026).