By Atoyebi Nike
The National Information Technology Development Agency has issued a new cybersecurity alert on emerging weaknesses found in ChatGPT, warning that the flaws could expose Nigerians to data leaks and manipulated AI responses.
Its Computer Emergency Readiness and Response Team said researchers uncovered seven vulnerabilities in the GPT4o and GPT5 models. The weaknesses allow attackers to exploit indirect prompt injection by hiding instructions inside webpages, comments or URLs that ChatGPT processes during browsing or summarization.
NITDA noted that malicious commands embedded in online content can force the model to perform unintended tasks, bypass safety controls or generate harmful responses. Some flaws also enable harmful inputs to remain in the system’s memory, influencing future conversations.
Although OpenAI has issued fixes, the agency said AI models still struggle to separate genuine user queries from hidden threats. It warned that compromised content could cause unauthorized actions, exposure of private data and long-term behavioral changes.
NITDA advised users and organizations to avoid browsing untrusted websites through ChatGPT, limit the use of memory features and keep deployed models updated to reduce risk.
