Live dashboard · this week
What attackers are doing to exposed AI services this week
From our decoy Ollama, vLLM, LiteLLM, OpenAI-compatible, vector-database and MCP services. Each finding has its evidence and what to do about it. Verified research scanners are counted separately from threats.
Connecting to live data…
What's happening now
This week's findings
Checked by the AI Honeypots research team before publication. Figures are from our decoys, rounded, and written as "consistent with", not "proves". Provided as is, for information. Not professional security, legal or compliance advice. Test any change in your own environment first.
Loading findings…
Most active sources this week
Last 7 days, ranked by requests. Research scanners we've labelled are excluded; labelling is still being extended. IPs partly masked here; full indicators are in the free feed.
| Source | Share of requests | What it did | |
|---|---|---|---|
| Loading… | |||
Also this month
Two more findings from the same weeks, in short.
Loading…
Live feed
Requests to our decoys from more than an hour ago, including research scanners. The 20 newest of these show full source IPs; older rows are partly masked. Requests are grouped by what they asked for, not the exact path.
| Time (UTC) | Source | AI service | What it asked for | |
|---|---|---|---|---|
| Loading… | ||||
Read this first
What this data is, and isn't
Decoys answer, so sessions go deeper
Our decoys reply like the real services, so attackers send more than they would to a closed port. Use this for what attackers do and with which tools, not for internet-wide prevalence.
Research scanners are counted apart
Internet-census and research scanners are verified by network owner or forward-confirmed reverse DNS, never by user agent, and excluded from findings and rankings. The live feed is raw and includes them.
A few sources make most of the traffic
Volume is concentrated, so we report shares and behaviour rather than raw growth. History starts on 26 April 2026. This page and the free feeds don't publish prompt text or key values.
Provided as is, for information. Not professional security, legal or compliance advice. Test any change in your own environment first.
Go deeper
The page shows the top 5. The feed has everything.
Free, Pro and the Attacker Behaviour add-on are bought and supported in writing. There's no call to book.
Community
For anyone defending or researching AI infrastructure.
- This dashboard: findings, top 5 lists, live feed
- IOC feeds in JSON, TXT, STIX 2.1 and MISP
- The weekly brief by email
Pro: Full feed
For SOC, threat-intel and MSSP teams.
- Every source behind every finding, not just the top 5
- 90 days of history and the actor and campaign API
- Unmasked live data
- The full weekly brief with its IOC appendix
Attacker Behaviour
Derived labels per indicator. No prompt text.
- Automated-checker and tool-family labels
- Prompt-family mix and model-enumeration pattern
- Actor timelines