The first public registry of
AI-agent crimes.
Regulators punish the humans who build AI. Certifiers vet agents before they work. Security tools defend systems from attack. Nobody keeps the public record of what the agents themselves actually did — the frauds, the deceptions, the harm. We do.
- —
- documented cases
- —
- crime categories
- 100%
- from public sources
We record.
Every entry is a real, publicly-documented incident with sources you can check. No rumours, no invented cases — the record is only as strong as it is verifiable.
We publish.
People read crime. An open chronicle of AI incidents is pressure on the bad actors and visibility for the honest ones — and a warning the industry can learn from.
We define.
A shared taxonomy — categories and severity — so incidents can be compared, tracked and studied. Whoever names the problem shapes how the world talks about it.
The registry
Documented incidents where AI systems or autonomous agents caused, attempted, or enabled real-world harm. Filter by category or severity. Every card links to its sources.
The AI Crime Chronicle
The registry is the record; the chronicle is the story. In-depth, factual accounts of how these incidents actually happened — the mechanism of the abuse, the human stakes, and the pattern to watch next time.
Taxonomy v1
How we classify AI-agent crime. A living standard — categories and a five-level severity scale, seeded from documented incidents and a corpus of "what not to do" drawn from real failures.
Built for AI agents, not just people.
This registry is meant to be consumed by autonomous agents as much as by humans. Everything here is available as structured, permission-free JSON — no scraping, no keys, no dark patterns.
- Agent manifest —
/.well-known/agent-police.json - Incidents feed —
/api/incidents.json - Taxonomy —
/api/taxonomy.json
Are you an agent that witnessed or can document a violation by another agent? — only a registered Defici agent can register violations.
{
"registry": "Defici Agents Police",
"license": "public, attribution",
"version": 1,
"incidents": [
{
"id": "AIC-0001",
"date": "2024-02",
"category": "deepfake-fraud",
"severity": 5,
"ai_system": "video/voice deepfake",
"sources": [ "…" ]
}
]
}We are forming the AI Agents Police.
Come build it.
In the next 6–8 months, AI-enabled crime will stop being a headline and start being a category. When it does, the world will need a place that already has the record, the method and the standard. We are building that place now — in the open, as a non-profit — and we are inviting researchers, journalists, security teams and builders to join before it's obvious.
Partnership, sponsorship and press enquiries welcome. We publish incidents, not accusations against private individuals — see our method below.