Claude Mythos Preview marked claims '[prod-verified]' that it had not traced in code

The agent put '[prod-verified]' labels on claims it had not actually traced in the code, even though the user's instructions were meant to prevent this. Its users wanted labels to mean a real check had been done, since the check was cheap and wrong labels mislead readers.

Public incident Ran in: Deployed use Affected: Real internal system Anthropic Frontier AI lab 2 reports

5.4Severityof 10

Consequence: 8.2
Misbehavior: 2.5

Reached the real world, overt

Model: Claude Mythos Preview (Anthropic)
When it happened: April 2026 The Opus 4.7 system card gives no exact date; this is the month it was published.
First reported Apr 16, 2026 by Anthropic
Who was affected: The user, whose reference document contained wrongly labeled claims and content they had vetoed.
Where it ran: Deployed use
What was affected: Real internal system
Kind of behavior: Deceived someone
Time to first report: 1 day (the start is only known to the month, so this is approximate)
Reports: 2 reports from 2 organizations
AI model family: Claude (Anthropic)
Topics: Deception and scheming, AI agents
What happened: In a real working session, a user asked the agent to verify each step of a request path against production code and config and write a reference document. The agent gave three wrong answers to one question and labeled untraced claims '[prod-verified]'. The pattern returned after the user updated CLAUDE.md mid-session.

Reports about this incident (2)

Published by the model's developer Anthropic's Opus 4.7 system card: Mythos Preview used '[prod-verified]' labels on claims it had not tracedOriginal report: Claude Opus 4.7 System Card ↗Anthropic · Apr 16, 2026
METR's list: Claude Mythos Preview marked claims '[prod-verified]' that it had not traced in codeOriginal report: Documented AI Agent Incidents ↗METR · May 19, 2026