Nigeria’s cyber‑governance watchdog has issued an urgent alert over security vulnerabilities in ChatGPT’s newest models, warning that the use of the artificial‑intelligence tool could expose unsuspecting users to serious data‑leakage and manipulation risks.
According to NITDA, models such as GPT‑4o and GPT‑5 harbour critical flaws that make them vulnerable to “indirect prompt‑injection” attacks, a method by which attackers embed hidden instructions into webpages, comments, or URLs, which the AI might execute inadvertently during normal browsing, summarisation, or search tasks.
These weaknesses reportedly allow misuse of the system to perform unauthorized actions, leak sensitive user data, produce manipulated outputs, or even “poison” the model’s memory potentially influencing future AI‑generated content in harmful or misleading ways.
In light of these risks, NITDA urged users especially businesses, government institutions and anyone handling sensitive data to disable ChatGPT’s browsing or memory features when interacting with untrusted web content, only enable advanced features when absolutely necessary, and ensure that deployed models are regularly updated and patched.
The advisory comes amid growing global concern over AI safety and serves as a stark reminder that even widely‑used tools like ChatGPT are not immune to security flaws and must be used with caution, especially in contexts involving personal or institutional data.
