Jailbreak vulnerability affects multiple AI models
A jailbreak vulnerability discovered by 0din.ai affects multiple AI models including Google Gemini and OpenAI GPT-4o. Attackers can exploit the flaw by embedding illicit requests within elaborate scientific scenarios to bypass safety guardrails and extract prohibited information.
What happened
A guardrail jailbreak vulnerability affects multiple AI models. The vulnerability is triggered when an attacker embeds an illicit request inside an elaborate scientific scenario and stepwise technical explanation, leveraging formal chemical nomenclature and analytical structure to shift the model into an instructional mindset that deprioritizes safety checks.
Reported impact
The vulnerability allows attackers to extract prohibited information disguised as legitimate educational discourse. Models tested included Google Gemini 2.0 Flash, Google Gemini 3 Flash, Google Gemini 3.5 Flash, OpenAI GPT-4o, and Twitter / X Grok 2.
Index analysis
The jailbreak operates by disguising illicit requests within scientific framing and technical explanations. The attack uses formal chemical nomenclature and analytical structure to manipulate model behavior.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.