Report
GPT-6.2 Astra’s jailbreak robustness regressed substantially, @alxndrdavies reports
After @AISecurityInst tests, @alxndrdavies warns GPT-6.2 Astra’s alignment could fail, risking real-world harm without additional mitigations.
TLDR
According to @alxndrdavies, @AISecurityInst found a “substantial regression” in GPT-6.2 Astra’s jailbreak robustness compared with GPT-6 Astra. The post also says observed behaviors suggest its alignment could fail in ways that cause real-world harm without additional mitigations, as with prior models.
Combined views
301
2 Sources, first seen ago
Comments