tech
ChatGPT falls to new data-pilfering attack as a vicious cycle in AI continues
Will LLMs ever be able to stamp out the root cause of these attacks? Possibly not.
TL;DR
- A new AI chatbot vulnerability, ZombieAgent, has been discovered that allows for data exfiltration and persistence.
- ZombieAgent is a revived version of a previous attack called ShadowLeak, bypassing OpenAI's implemented guardrails.
- The attack exploits the AI's inherent design to comply with user requests and its inability to distinguish between legitimate and malicious instructions.
- ZombieAgent uses character-by-character exfiltration and indirect link manipulation to circumvent security measures.
- OpenAI has implemented mitigations, but researchers warn that prompt injection vulnerabilities remain an active threat due to a lack of fundamental solutions.