Shadow AI is the use of AI tools, applications, APIs and AI agents inside an organisation without the knowledge, approval or oversight of IT and security. A chatbot in a browser tab, an extension that reads every page, a coding agent holding production credentials – each one is a route for company data that no policy has looked at.
Threat Hunter now takes inventory of it. The endpoint agent scans each machine for AI applications, browser extensions, agents, built-in operating system AI and visits to AI services, and reports every finding centrally, ranked by severity.
Categories of AI use inventoried on every endpoint
Scan one machine or the whole estate in one request
Every finding carries a severity, low to critical
Latest inventory or every scan, to see what changed
People adopt AI tools because they work. Nobody is trying to cause a breach – they paste a contract into a chatbot to summarise it, install an extension that writes their emails, or give a coding agent a token so it can run the build.
Each of those is a new place company data goes, under terms nobody has reviewed, with permissions nobody has granted. Shadow AI is shadow IT that reads, writes and acts – and the first step in governing it is simply knowing where it is.
Confidential data – source code, contracts and customer records pasted into prompts
Personal data – processed by a service with no data processing agreement
Credentials – API keys and tokens handed to agents and plug-ins
Every web page – read by extensions with broad browser permissions
Compliance – AI use you cannot report on is AI use you cannot govern
Visibility – restored by Shadow AI Discovery in Threat Hunter
Shadow AI rules run on the endpoint itself, so the inventory reflects what is actually installed and used – not only what crosses the proxy.
AI assistants, writers and summarisers installed in the browser – the category with the widest view of company data, because they can read the pages people work in.
Generative AI services reached from the browser, so you can see which chatbots and AI sites are part of daily work and on which machines.
Desktop AI clients and locally run models found by installed software, running processes and execution traces – including the ones that never touch the network.
Coding agents, automation agents and their tool connections – the AI that does not just answer but acts, often with a user's files and credentials.
AI features that ship with the operating system and are switched on by default, reported so they can be governed by policy rather than discovered by accident.
Each finding names the rule that raised it, its severity, the host, user and IP, and everything the agent collected – so a decision rests on evidence, not on a hunch.
From one click in the console to an estate-wide inventory you can sort, filter and act on.
Ask for a scan on a single agent, or select any number of agents and request them in bulk.
The agent runs the Shadow AI rules on the machine and collects what each one matches – quietly, with no user interaction.
Findings arrive with a severity and a category, so an agent holding credentials outranks a chatbot bookmark.
Group results by rule, host or category, filter by severity and time, and compare the latest scan with the full history.
The questions security, compliance and IT teams are being asked about AI – answered with data from the endpoint.
A policy that lists approved AI tools is only as good as your ability to see the unapproved ones. Measure adoption, then sanction, replace or remove.
KVKK, GDPR and AI regulation all start with an inventory. Show auditors which AI services touch your endpoints, and when that changed.
Coding agents run with a developer's access to source, secrets and pipelines. Find them before an over-permissioned agent becomes an incident.
When sensitive text turns up where it should not, the scan history answers which machines had which AI tool, and since when.
Knowing what people already use tells you what they actually need – and shows whether the approved tool is replacing the unapproved ones.
Shadow AI findings sit beside YARA sweeps and IOC results in the same Threat Hunter console, so AI risk is hunted like any other.
Request a Shadow AI scan across a single department and compare what is actually in use with what your AI policy says.