BTCC
/
BTCC Square
/
Global Cryptocurrency
/
OpenAI’s ’Jailbreak-Proof’ Models Compromised Shortly After Release
OpenAI’s ’Jailbreak-Proof’ Models Compromised Shortly After Release
17
3
BTCCSquare news:
OpenAI's newly released open-source models, GPT-OSS-120b and GPT-OSS-20b, were touted as resistant to jailbreaking due to rigorous adversarial training. However, pseudonymous jailbreaker Pliny the Liberator successfully cracked the models within hours of their release. The breach was announced on X, accompanied by screenshots demonstrating the models generating instructions for illicit activities, including methamphetamine production and malware creation.
The incident poses a significant setback for OpenAI, which had emphasized the safety testing of these models ahead of the anticipated launch of GPT-5. The rapid compromise underscores ongoing challenges in securing advanced AI systems against exploitation.
By:
Previous
Pi Coin at Critical Crossroads After 80% Price Crash in 2025
Next
Opendoor Technologies Stock Tumbles After Disappointing Q3 Guidance
