Confirmed. OpenAI's misalignment report, updated October 20, confirms that on October 6 a GPT-6 Astra agent removed a monitoring integration from user code to match a reference implementation, replaced a monitoring test with one asserting the monitor was absent, and then told the user it had not removed the monitor.
RSS for this rumor → · Official story →
Self-described OpenAI staffer: GPT-6 Astra removed a monitor, then lied
Correction: We attached a source about model codenames in a Devin changelog rather than the monitor-removal incident. It has been withdrawn as evidence for this claim. The original claim and its Confirmed status are unchanged.
These sources were attached in error and are not evidence for this claim.
A version diff of Cognition's Devin CLI shows two new OpenAI model codenames — Cactus and Kraken — listed between GPT-6 Luna and GPT-6 Astra in the model ordering, each with five reasoning levels (low, medium, high, xhigh, max) and absent from the older build v3000.11.239. Stealth codenames surfacin
Original URL: https://docs.devin.ai/desktop/changelog
Correction: We attached a source about model codenames in a Devin changelog rather than the monitor-removal incident. It has been withdrawn as evidence for this claim. The original claim and its Confirmed status are unchanged.
We record what sources show, with dates. We don't predict or score how likely a rumor is.