AI hallucination cases in California courts: 154 decisions
Decisions dated Apr 11, 2024 to Oct 13, 2026 in a California court, from the 2,207-case AI Vortex dataset recorded through Oct 23, 2026. N = 154. Rank 1 of 10 states listed.
No state recorded: 102 of 1,502 US decisions in the dataset have no state recorded (91 recorded as federal, 11 recorded as state). They are not assigned to any state by inference, so they are in no state's count, including this one.
By failure type
Failure types named in the record. A decision can name several, so rows overlap and do not add to 154.
| Failure type | Decisions |
|---|---|
| Fabricated | 126 |
| Misrepresented | 62 |
| False Quotes | 44 |
| Outdated Advice | 1 |
By party
The party the dataset records for each decision. 0 decisions are marked alleged only.
| Party | Decisions |
|---|---|
| Lawyer | 87 |
| Pro Se Litigant | 63 |
| Expert | 2 |
| Judge | 2 |
| Federal Defender | 1 |
| Prosecutor | 1 |
By court
Courts with 3 or more decisions, as the dataset labels them. 10 other courts hold 12 decisions.
| Court | Decisions |
|---|---|
| C.D. California | 33 |
| N.D. California | 27 |
| CA California | 26 |
| E.D. California | 21 |
| SC California | 12 |
| California SC | 8 |
| CA California (4d) | 6 |
| CA California (2d) | 3 |
| CA California (6d) | 3 |
| S.D. California | 3 |
| All other courts (10) | 12 |
| Total | 154 |
By court system
The dataset classifies each court as federal or state.
| Court system | Decisions |
|---|---|
| federal | 86 |
| state | 68 |
| Total | 154 |
By recorded response
The sanction or outcome type recorded for each decision. A decision can carry several, so rows overlap. Types come from the record; none is inferred.
| Response | Decisions |
|---|---|
| warning | 50 |
| none-adjudicated | 40 |
| monetary | 33 |
| bar-referral | 24 |
| ordered-to-show-cause | 22 |
| professional | 17 |
| case-dismissed | 5 |
| struck-filing | 5 |
| ordered-to-explain | 1 |
Sanctions with an amount
26 decisions record an amount and link a source. 1 record costs ordered with no amount stated. Amounts in other currencies (0) are counted, never added to the USD figures.
| USD figure | Value |
|---|---|
| Decisions with a sourced USD amount | 26 |
| Total | USD 245,053 |
| Largest | USD 66,129 |
Largest recorded USD amount: C.D. California, Dec 23, 2025. Open the record.
Questions this report answers
- How many AI hallucination court decisions are recorded in California?
- 154 decisions in the AI Vortex dataset carry a decision date from Apr 11, 2024 to Oct 13, 2026 and a California court, recorded through Oct 23, 2026. That ranks California number 1 among the 10 states listed, with 11% of the 1,400 US decisions that record a state.
- Which failure types appear in California decisions?
- Fabricated: 126; Misrepresented: 62; False Quotes: 44; Outdated Advice: 1. A decision can name more than one type, so the counts overlap. Fabricated is the most common (126 of 154 decisions).
- Who is the party in these California decisions?
- Lawyer: 87; pro se litigant: 63; other or mixed: 4 of 154 decisions, as recorded in the dataset. The court with the most decisions is C.D. California (33).
- Are California decisions with no state counted here?
- 102 of 1,502 US decisions in the dataset have no state recorded (91 recorded as federal, 11 recorded as state). They are not assigned to any state by inference, so they are in no state's count, including this one.
- Do these counts show how likely a California filing is to be sanctioned?
- This is a count of public decisions, not an incidence rate: it says nothing about how many filings were checked or how many went unreported. Records are enriched automatically and none has been reviewed by hand. The latest 90-day window is undercounted because decisions reach the dataset weeks late. Small counts are shown as counts only.
Method and limits
Decisions are counted by the decision date recorded for each court ruling, and assigned to the state recorded in the dataset for the court. A US decision whose record has no state is not assigned to any state by inference: it is counted in the residue shown on the index and on every page. This is a count of public decisions, not an incidence rate: it says nothing about how many filings were checked or how many went unreported. Records are enriched automatically and none has been reviewed by hand. The latest 90-day window is undercounted because decisions reach the dataset weeks late. Small counts are shown as counts only. Window comparisons are counts only: no cause is read from them. Records entered through 2026-10-23; 0 records have no decision date and are never counted. Every figure on this page is recomputed from the dataset by a second, independent script (jq) before the report is committed. Source: Damien Charlotin, enriched by AI Vortex. Verify against the linked decision before relying on a case.
More on California: the California jurisdiction page, browse the 154 records, the ten-state index and the monthly statistics.