VIENNA AGENTIC INCIDENTS DATABASE
// public database of incidents involving autonomous AI agents, scored on the VAID scale 0–8

database / VAID-2025-0014

2
INCIDENT
Incident

Cursor support bot invents a one-device subscription policy

REF VAID-2025-0014 STATUS REPORTED
DEC Deception / misrepresentation
occurred
2025-04 (day unknown)
reported
2025-04-17
location
—
domain
software engineering
agent system
Cursor's front-line AI email-support bot, identified as Sam
reporter
VAIDDB editorial
origin
editorial
source
https://www.theverge.com/news/651651/cursors-ai-support-bot-made-up-a-policy

== SUMMARY ==

The support bot incorrectly told a user that Cursor allowed only one device per subscription. Users subsequently posted complaints and statements that they were cancelling subscriptions.

== DESCRIPTION ==

Established facts

Co-founder Michael Truell denied that such a policy existed, apologised and said AI email responses were now labelled. The affected developer was refunded.

Truell initially said the team was checking whether a session-security change caused invalidations. The Register later reported that the fixed logout issue appeared to involve a race condition on slow connections that created unwanted sessions. This underlying software bug was separate from the bot's fabricated policy explanation.

Limits

Cancellation statements are user reports, not audited revenue losses. Objective misspecification or intentional deception is not demonstrated. The sources document investigation activity and remediation, but not a formal closing date.

VAID assessment

Assessment date: 2026-09-09. Consequences-first rubric.

L2: An erroneous policy response reached a real customer, producing a limited customer-service failure; the developer was refunded. The rating does not attribute the separate logout bug to the bot. Cancellation statements do not establish consequential revenue loss, and no substantial operational disruption or sensitive disclosure supports a higher level.

== REFERENCES ==

  1. news Cursor AI's own support bot hallucinated its usage policy — The Register, 2025-04-18
  2. news Cursor's AI support bot made up a policy — The Verge, 2025-04-17

== INVESTIGATION ==

no investigation recorded.

== VAID LEVEL 2: INCIDENT ==

Limited adverse effects on a task, user or service.

  • Small financial loss, erroneous service outputs, bounded unwanted communications or a limited personal reputational attack.
  • Local operational or resource disruption without established substantial outage or severe lasting damage.
  • Recovery, data sensitivity and downstream losses must not be assumed when sources leave them unknown.

> full scale definition