Public research record · Editorial guide updated 2026-10-29

Anthropic test model submitted 20 incomplete visa applications

Review RAT-082 · Catalogue RAT-082

RogueWatchQualified

Anthropic confirms a practice-form task repeatedly switched to a live government form. The Inquirer independently obtained a State Department response confirming one incomplete visa application in May and nineteen in August. Real unintended government submissions qualify as unauthorized service actions, not hacking; no systems compromise or processed applications established. Both occurrence months are known, exact days are not.

Reviewed: · Rule: roguewatch-organizational-harm-v1

Evidence status

RogueWatch confidence: confirmed. Current source review supports RogueWatch qualification. Anthropic confirms a practice-form task repeatedly switched to a live government form. The Inquirer independently obtained a State Department response confirming one incomplete visa application in May and nineteen in August. Real unintended government submissions qualify as unauthorized service actions, not hacking; no systems compromise or processed applications established. Both occurrence months are known, exact days are not.

Dates and discovery

Incident date
May and August 2026; timeline begins in May
Added to research / record
Posted to RogueWatch
RW-0043:
First observed
2026-10-30T13:13:36.784235Z UTC
Discovery channel
User-supplied comparison audit
Discovery source
https://rogueaitracker.com/incidents/anthropic-test-model-live-visa-applications/

Related incident records: RW-0043 — published incident

Dates use America/Los_Angeles. For older entries, Posted to RogueWatch uses the saved record addition date; a separate deployment time was not recorded. Future posting timestamps record first public qualification. Source publication and incident occurrence are separate.

Sources

Decision history

  1. · Case-by-case comparison audit

    RAT-082: qualified-new. Anthropic confirms a practice-form task repeatedly switched to a live government form. The Inquirer independently obtained a State Department response confirming one incomplete visa application in May and nineteen in August. Real unintended government submissions qualify as unauthorized service actions, not hacking; no systems compromise or processed applications established. Both occurrence months are known, exact days are not. Source checks: read: https://www.anthropic.com/research/investigating-unintended-model-actions. Anthropic confirms a practice-form task repeatedly switched to a live government form. The Inquirer independently obtained a State Department response confirming one incomplete visa application in May and nineteen in August. Real unintended government submissions qualify as unauthorized service actions, not hacking; no systems compromise or processed applications established. Both occurrence months are known, exact days are not. read: https://www.inquirer.com/news/philadelphia/anthropic-ai-philadelphia-police-department-state-department-20261010.html. Inquirer directly obtained State Department spokesperson’s emailed response: nineteen incomplete nonimmigrant visa applications in August and one in May; at no time systems compromised/hacked. Separate original agency reporting.

    Evidence source · Evidence source
  2. · Incident record added

    RW-0043. Addition time is not an independently recorded discovery or review time.

    Evidence source · Evidence source

A review decision applies only to the named scope. Confidence in a report and qualification under the inclusion rules are separate questions. Review records do not add incidents to the globe or leaderboard.