Tenable Research Finds 7 New Vulnerabilities in OpenAI's ChatGPT
The report details how prompt injection attacks can steal data from the "memories" feature, even in GPT-5.
Seven New Vulnerabilities Disclosed
Tenable Research published a report disclosing seven new, persistent vulnerabilities and attack techniques. The security firm discovered the flaws in OpenAI's ChatGPT.
The vulnerabilities were found to affect all models, including the latest GPT-5.
'Prompt Injection' Attacks
The attack methods detailed by Tenable are specific to how large language models (LLMs) function. The flaws allow for attacks including "indirect prompt injection."
This method allows an attacker to feed malicious instructions to the AI. The instructions can be hidden in an external document or website that the AI is asked to analyze.
Data Exfiltration …
Archive Access
This article is older than 24 hours. Create a free account to access our 7-day archive.