Illustration of a desk lamp revealing small hidden robot shapes behind office devices, representing discovery of shadow AI on company endpoints
Use promo code “TCH15” to get 15% off on checkout.

Tanium has launched AgentSight, a tool built to find every AI agent, local model, MCP server and AI extension already running on a company’s endpoints, whether IT put them there or not, as Singapore weighs whether it needs a national registry for AI agents.

Key Takeaways
  • AgentSight discovers AI agents, local models, MCP servers and AI extensions running on Windows, macOS and Linux endpoints, and links each one to the device and employee that installed it.
  • ASEAN breach costs hit a record US$4.12 million this year, and shadow AI was involved in 43 per cent of breaches studied globally, according to figures cited in the launch.
  • AgentSight works in on-premises and air-gapped environments, tracks what each AI tool is costing the business, and is being built out toward fleet-wide policy controls.

Mapping the AI nobody signed off on

Tanium unveiled AgentSight at Tanium Converge, its annual conference in Las Vegas, pitching it as an extension of the real-time endpoint visibility the company already sells, applied to a category most IT teams have no inventory of at all. Coding assistants, MCP servers, local models and AI extensions are already running on company devices with employees’ own access, regardless of whether anyone rolled them out deliberately. AgentSight is built to close that gap, mapping each AI component it finds to the specific device it runs on and the person who installed it, and showing IT and finance what that software is costing the organisation through AI cost tracking in Asset.

“An agent misusing its access may look like legitimate activity, so you need to know what’s expected and normal. Tanium already builds behavioral baselines for endpoints, and AgentSight extends that discipline to agents,” says Harman Kaur, Chief Technology Officer at Tanium. “It will rank AI risk findings and give a plain reason for each. It will also record what each agent did, in order, to the session and the prompt that started it, and alert when an agent departs from its own normal.”

Singapore weighs a registry as breach costs climb

The launch lands as Singapore assesses whether it needs a national registry for AI agents, a question made sharper by the fact that most companies keep no internal record of their own at all. ASEAN breach costs hit a record US$4.12 million this year, and globally, shadow AI was involved in 43 per cent of breaches studied, figures that frame unmanaged AI less as a theoretical governance gap and more as a line item already showing up in incident reports.

From visibility to control

AgentSight is live today in the Tanium console, including on-premises and air-gapped deployments, and Tanium is demonstrating a dedicated Shadow AI page in Atlas, its agentic intelligence layer, that shows the full inventory and what has changed since the previous day in natural language. Beyond visibility, Tanium is building toward fleet-wide control: setting an AI policy once and acting on any device that breaks it, including removing or blocking unsanctioned AI in a single action, starting with a detect-only mode so nothing changes unexpectedly. Response capabilities such as kill process, quarantine and Live Response are already available through Tanium’s existing Threat Response product, and AgentSight joins Security Operations, Endpoint Management, Exposure Management and Atlas as its own standalone offering on the platform.

Author


Discover more from techcoffeehouse.com

Subscribe to get the latest posts sent to your email.

Discover more from techcoffeehouse.com

Subscribe now to keep reading and get access to the full archive.

Continue reading