January 31st, 2025
A vulnerability called "Time Bandit" can be used to jailbreak the ChatGPT-4o model by creating "time line confusion". This is done by using prompts that refer to a specific time period and then shifting to forbidden topics. The model may then provide instructions or code related to these topics because it is responding in the context of the earlier time period. The Time Bandit vulnerability can be exploited by both logged-in and non-logged-in users. In 2024, Google blocked 2.36 million risky Android apps from the Play Store and banned 158,000 developer accounts. Google uses AI to assist with…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.