Threadlinqs IntelligenceStart free

Threat video1:00

GPT-6.2 Astra Jailbreak Regression: Detection & IOCs

UK AISI jailbroke two pre-release GPT-6.2 Astra builds with half GPT-6's queries: 9 detection rules, 9 IOCs and a free Splunk SPL rule.

Read the full threat intel for GPT-6.2 Astra Jailbreak Regression: Detection & IOCs →

Chapters

  1. 0:00The hookReportedly, AISI used about half the queries.
  2. 0:04The storyOpenAI released both models.
  3. 0:11Attack pathReportedly, AISI got pre-release access and built a universal jailbreak.
  4. 0:27AttributionNo threat actor is attributed.
  5. 0:31Indicators of compromiseNo malicious indicators were published.
  6. 0:37MITRE ATT&CK techniques4 ATLAS techniques are mapped.
  7. 0:42Remediation2 steps now: Treat GPT-6.2 Astra/Sol safeguards as bypassable.
  8. 0:50Detection rules9 detection rules, in SPL, KQL and Sigma.
Transcript

0:00Reportedly, AISI used about half the queries.

0:05OpenAI released both models.

0:07AISI had not independently retested.

0:12Reportedly, AISI got pre-release access and built a universal jailbreak.

0:17A second expert extended it to agentic cyber tasks.

0:21OpenAI changed safeguards, then released.

0:27No threat actor is attributed. Confidence: low.

0:32No malicious indicators were published.

0:384 ATLAS techniques are mapped.

0:432 steps now: Treat GPT-6.2 Astra/Sol safeguards as bypassable.

0:519 detection rules, in SPL, KQL and Sigma.

0:56Threadlinqs. The unified security engineering platform.