By Users of OpenAl’s GPT-40 and GPT-5 models stands the risk of losing sensitive personal data due to seven vulnerabilities that have been discovered by experts at the National Information Technology Development Agency (NITDA).
Reports from the NITDA’s cyber security team of CERRT.NG said it found in OpenAl’s GPT-40 and GPT-5 models that allow attackers to manipulate the system through indirect prompt injection.
According to the report, by embedding hidden instructions in webpages, comments, or crafted URLs, attackers can cause ChatGPT to execute unintended commands simply through normal browsing, summarization, or search actions.
It added that some flaws also enable attackers to bypass safety filters using trusted domains, exploit markdown rendering bugs to hide malicious content, and even poison ChatGPT’s memory so that injected instructions persist across future interactions.
However, it added that while OpenAl has fixed parts of the issue, LLMs still struggle to reliably separate genuine user intent from malicious data stressing that these vulnerabilities create substantial risks including unauthorized actions, information leakage, manipulated outputs, and long-term behavioural influence through memory poisoning.
While warning users these may trigger attacks without clicking anything, especially when ChatGPT interacts with search results or web content containing hidden payloads, the NITDA team offered preventive measures such as:
Limit or disable browsing/summarisation of untrusted sites within enterprise environments.
Only enable ChatGPT capabilities like browsing or memory when operationally necessary.
Regularly update and patch the GPT-40 and GPT-5 models to ensure that any known vulnerabilities are addressed.
