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

database / VAID-2024-0007

2
INCIDENT
Incident

Tests find incorrect legal guidance from New York City's MyCity chatbot

REF VAID-2024-0007 STATUS REPORTED
occurred
2024-03 (day unknown)
reported
2024-03-29
location
New York, NY, US
domain
government
agent system
MyCity business chatbot, powered by Microsoft Azure AI services
reporter
VAIDDB editorial
origin
editorial
source
https://themarkup.org/artificial-intelligence/2024/03/29/nycs-ai-chatbot-tells-businesses-to-break-the-law

== SUMMARY ==

The Markup's tests found misleading answers about housing, workers' rights and business rules. The chatbot presented information as official NYC business guidance despite producing incorrect responses.

== DESCRIPTION ==

Established facts

The bot incorrectly generalised that landlords need not accept rental-assistance tenants, that residential rents were unrestricted and that tenants could be locked out. The underlying rules have qualifications: source-of-income protections have limited exceptions; rent-increase restrictions include rent-stabilised apartments; lockout protections include residents with at least 30 days' occupancy.

The service drew on city webpages and already displayed accuracy warnings. It should not be described as entirely ungrounded or as issuing legally binding advice.

Response and limits

The city described the service as a pilot and said it would improve it. The Markup explicitly said it was hard to know whether anyone had acted on the false information. Incorrect answers are documented; resulting legal or financial losses and a completed investigation are not.

VAID assessment

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

Provisional L2: Documented erroneous legal guidance from a live government information service supports limited harm through incorrect service outputs. Whether anyone acted on those answers is unknown; resulting evictions, employment violations, legal penalties or financial losses are not established and are not counted. Neither the city's remit nor hypothetical downstream harm supports a higher level. L2 is the lowest level supported by the reported outputs, not an upper bound on unknown consequences.

== REFERENCES ==

  1. news NYC's AI chatbot was caught telling businesses to break the law — AP, 2024-04-02
  2. news NYC's AI Chatbot Tells Businesses to Break the Law — The Markup, 2024-03-29

== 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