hackingincidents.ai
Incidents289Sources240Dataset2026-11-17Last sweep2026-11-17
Ledger / Data
HI // Data

The data

The whole ledger as files you can download, with a guide to every field.

Incidents289one row each
Sources271registered, with role and kind
Last updated17 Nov 2026we aim to refresh daily

The whole ledger is free to download. It is current to 17 Nov 2026 and holds 289 incidents. It is a list of published reports and it will have gaps and errors; the about page says what it can and cannot tell you.

Incidents (JSON)

Incidents with their grade, reasons, models, enrichment and sources.

/incidents.json

289 incidents · 654 KB

Sources (JSON)

Every registered source with its kind, role and sweep date.

/sources.json

271 sources · 44 KB

Incidents (CSV)

One row per incident, for a spreadsheet. List fields are joined with semicolons.

/incidents.csv

289 incidents · 222 KB

Sources (CSV)

One row per registered source, with the number of incidents that cite it.

/sources.csv

271 sources · 23 KB

Field guide

The fields in incidents.json, with the values a closed field can take. The sources file has two closed fields of its own: role (swept, cited, unreachable); kind (model-provider, vendor-ti, government, court-record, incident-database, academic, press, regional-cert). Source rows also carry swept (the sweep date) and, for sources we could not read, public_note.

id
The incident's address: /incident/<id>/.
title
A one-line statement of what happened.
date
When the activity happened, as YYYY-MM-DD or YYYY-MM; the disclosure date when the source gives no other.
disclosed
When it was made public, where that differs from the activity.
activity
When the activity ran, where the source gives a span.
actor
The actor's name as the source gives it. This site makes no attribution of its own.
actor_type
What kind of actor the source says it was.Allowed values: criminal, nation-state, hacktivist, unknown
nexus
The country or region the source ties the actor to, if any.
targets
Who was hit, as the source describes it.
ai_role
What the AI did. One or more of the values below.Allowed values: recon, vuln-research, exploit-dev, malware-dev, malware-runtime-llm, phishing-content, social-engineering, deepfake, ops-automation, agentic-intrusion, data-analysis, evasion, fraud, supply-chain, llmjacking, agent-hijack
models
The models or tools the source names, as named. A placeholder such as "Unnamed LLM" means the source gave no name.
model_basis
Where the model name comes from.Allowed values: stated-by-vendor, stated-by-researcher, stated-by-victim, inferred, unknown
autonomy
How much of the attack the AI ran.Allowed values: assist, semi-autonomous, autonomous
outcome
What the operation achieved.
confidence
The grade: how strong the evidence of AI use is. See the method page.Allowed values: CONFIRMED, STRONG, SUSPECTED
confidence_reason
Why that grade.
summary
A short plain-words account.
quote
A sentence from one of the incident's own sources that supports the grade.
single_source_reason
Present when fewer than two independent publishers report the incident: why that is the case.
sources
The incident's sources, each with the fields below.
sources[].title
The source's own headline.
sources[].publisher
Who published it.
sources[].url
Where it is.
sources[].date
When the source was published.
sources[].kind
A reading note from extraction; the registry kind in sources.json is the one to use.
sources[].verified_on
The date we last fetched the source. It records that the page answered, not that its claims were checked.
sources[].quote_found
True when the incident's supporting quote was found in the text of this page.
sources[].archive_url
An archived copy, where one exists.
sources[].fetch_status
The HTTP status of the last fetch.
sources[].primary
True when this is a primary source for the incident: the provider, the victim, a government body, a court record, or the investigating firm.
sources[].primary_via_party
Set when the source is a named party's on-record statement quoted by an outlet: the party.
sources[].aggregator_only
True when the sources cited are database records only.
enrichment
Derived facts, each with the fields below.
enrichment.models
Each model with its name, vendor, openness and kind.Allowed values: openness: closed, open-weight, criminal, unknown, kind: llm, coding-agent, deepfake-video, voice-clone, image, harness, other
enrichment.model_named
Whether the source names a model at all.Allowed values: true, false
enrichment.harness
Whether an agent harness was used, which ones, and what kind.Allowed values: kind: open-source, commercial-agent, custom, none
enrichment.countries
The attacker's country as the source attributes it, as ISO 3166-1 alpha-2 codes.
enrichment.country_basis
How the country was reached.Allowed values: attributed, language-or-infrastructure, none
enrichment.sectors
The sectors hit, one to three.Allowed values: finance, crypto, government, defense, technology, ai-platforms, software-supply-chain, telecom, healthcare, education, critical-infrastructure, energy, manufacturing, retail, media, legal, individuals, enterprise-broad, unknown

License

The data is published under the license of the Substrate repository: MIT License, Copyright (c) 2025 Daniel Miessler. Short quotes from sources are credited to their publishers and remain under the publishers' terms. If you use the data, please cite it as hackingincidents.ai, Unsupervised Learning, accessed <date>.