AI Hacking Incidents
An attempt to track cyberattacks where attackers used AI, from published reports. We will have missed some and gotten some wrong; we aim to refresh it daily, and corrections are welcome.
289 reported attacks, graded by evidence
Each entry is one operation that a model provider, a security firm, a government or the press published, where AI use is stated or credibly suspected. The grade says how strong that evidence is. These are counts of published reports, so they show what got noticed and written up, which is not the same as what happened.
- Confirmed
- The model provider, the victim, a government, or the primary investigator states that AI was used in this operation, with evidence such as provider telemetry, recovered prompts, a recording, forensic detail, or malware that calls an LLM.
- Strong
- A credible security firm or agency attributes AI use from artifacts, such as LLM-written code or comments, but the attribution is an inference rather than direct observation.
- Suspected
- The case is reported and AI use is plausible, but the evidence is thin, comes from one secondary source, rests on an attacker's own claim, or reaches us only second-hand.
- Method
- How each grade is earned
2023 52024 272025 802026* 177
* 2026 to 17 Nov 2026
146 confirmed110 strong33 suspected
On its ownSemi-autonomousAssisted
119 named170 not named
57 with two or more201 with one independent publisher31 known only through an incident database
Counts published reports. An attack nobody wrote up is not here.
The record, month by month
One cell per incident in the ledger, placed in the month the ledger dates it to. Colour is how much of the attack the AI ran; tap a cell for the incident.
Latest entries
The most recent incidents in the ledger.
The rest of the site
Where to go from here.