The National Information Technology Development Agency (NITDA) has issued a security alert warning Nigerians about newly discovered vulnerabilities in OpenAI’s GPT-4.0 and GPT-5 series, which could expose users to data leakage and other cyber risks
The warning was contained in a statement released on Monday by NITDA’s Director of Corporate Affairs and External Relations, Hadiza Umar, in Abuja.
According to her, the agency has identified seven critical weaknesses in the models, exposing users to attacks through indirect prompt injection—a method that allows hidden commands to be planted in webpages or online content.
Her words: “By embedding hidden instructions in webpages, comments or crafted URLs, attackers can cause ChatGPT to execute unintended commands through normal browsing, summarisation or search actions.”
She added that some of the vulnerabilities also make it possible for attackers to bypass safety filters using trusted domains and exploit markdown rendering bugs to conceal harmful content.
That act can even poison ChatGPT’s memory so that injected instructions persist across future interactions,” Mrs Umar warned.
She noted that although OpenAI has begun addressing some of the flaws, large language models still struggle to differentiate genuine user intent from malicious embedded data.
Reinforcing the agency’s concern, she explained that the technique involves planting invisible instructions in webpages, online comments, or specially crafted URLs — misleading ChatGPT into performing actions without the user’s knowledge.
Mrs Umar said these vulnerabilities carry significant risks, including unauthorised actions, information leakage, manipulated outputs, and long-term behavioural influence due to memory poisoning.
To mitigate the dangers, she urged organisations to limit or completely disable ChatGPT’s browsing and summarisation of untrusted websites within enterprise systems.
“Only enable ChatGPT capabilities like browsing or memory when operationally necessary,” she cautioned.
Mrs Umar also advised users and institutions to regularly update GPT-4.0 and GPT-5 models to ensure that all known vulnerabilities are adequately patched.
