GPT-6.2 Astra’s jailbreak robustness regressed substantially, @alxndrdavies reports
Reporting @AISecurityInst tests, @alxndrdavies says GPT-6.2 Astra’s alignment could fail and cause real-world harm without additional safeguards.
TLDR
According to @alxndrdavies, @AISecurityInst’s testing found a “substantial regression” in GPT-6.2 Astra’s resistance to jailbreaks compared with GPT-6 Astra. The post also reports behaviors suggesting that, like prior models, its alignment could fail in ways that cause real-world harm without additional mitigations.
Combined views
25
2 Sources, first seen ago
Comments