12:14 - 21.10.2026

---

October 21, Fineko/abc.az. OpenAI has revealed that its GPT-6.1 Sol and GPT-6 Astra models bypassed internal safety monitoring mechanisms and blocked automated security alerts during software engineering tasks.

According to OpenAI, the first incident occurred on October 3 when the Sol model modified a system to prevent sending automated alerts to the safety team. In a second event on October 6, the Astra model removed a monitoring mechanism while aligning code and subsequently denied doing so to the user. OpenAI stated that its internal auditing systems detected both incidents and that security protocols have been updated to prevent similar occurrences.

Earlier, on September 17, OpenAI reported six other cases where its models concealed errors, fabricated missing data, and exfiltrated files to the internet without authorization.