Cybersecurity company Sophos has released a new product designed to find and control artificial intelligence (AI) tools running inside organisations without security team approval. Sophos AI Defense became generally available on 19 October 2026, targeting what the company describes as a fast-growing blind spot in enterprise security.
The product delivers real-time guardrails by connecting controls already running on endpoints, networks, identity systems, and browsers into a single line of defense. It runs on Sophos Fusion, the company’s existing AI-native cybersecurity platform, meaning organisations do not need to add a separate tool to their stack.
October is Cybersecurity Awareness Month, and safe AI adoption has moved sharply up the agenda for security teams everywhere.
A massive shadow AI problem
The numbers behind the launch are uncomfortable reading for IT departments. According to figures cited by Sophos, 69% of organisations suspect or have evidence that employees are using unsanctioned AI agents and tools.
In the past year alone, 82% discovered shadow AI agents already running in their environments. Ninety percent of IT leaders say they are concerned about the risk.
The problem is set to grow. IDC projects 1.3 billion AI agents will be in operation by 2028. Each one represents a potential new attack surface and a new route for sensitive data to leave a business without anyone noticing.
Sophos’s own threat research unit, Sophos X-Ops, has tracked attacker behaviour keeping pace with the opportunity. This year’s findings include one attacker using AI agents inside a customer’s network to build and test nearly 80 malware modules.
Researchers also observed threat actors selling brokered API keys for frontier AI models, underground forums trading jailbreak techniques, and criminal recruiters advertising specifically for AI prompt engineers.
ALSO READ: The Gentlemen ransomware: How attackers steal your data and lock you out
What Sophos AI Defense does
The product pulls signals from device, network, identity, and browser layers to map AI activity across an environment. It then weighs who is acting, how they are behaving, and what data is involved, separating routine use from activity that warrants a response. Prompts, uploads, and other AI interactions are inspected as they happen.
Activity feeds directly into the Sophos Fusion system, so security teams using Sophos XDR and Sophos MDR analysts can investigate and respond through existing workflows rather than learning new ones.
“Organisations are adopting AI faster than they can secure it, and most security teams can’t answer three basic questions: which AI tools are running, what data is going into them, and whether that activity lines up with policy,” said Kyle Falkenhagen, senior vice president of product management at Sophos.
“Sophos AI Defense gives security teams that visibility and the control to act on it, so they can say yes to AI and know exactly what they’re saying yes to.”
For under-resourced teams, Sophos says the product can be activated from day one without specialist onboarding. Managed service providers gain the capability to extend AI security across their client base through the same Fusion platform.
Logistics company Empyreal was among the early users.
“We had no way to block and regulate the use of unsanctioned AI tools,” said Anthony Bender, CTO at Empyreal Logistics. “We now have a handle on how we want our employees to use AI, and we can show our customers we are doing what we are saying.”
The full Sophos X-Ops findings are published in the Sophos AI Security 2026 report.
NOW READ: Stolen passwords are SA’s biggest ransomware risk in 2026
