NITDA Warns of AI Security Risks in OpenAI Models

The National Information Technology Development Agency (NITDA) of Nigeria has issued a warning regarding security vulnerabilities in OpenAI's GPT-4 and GPT-5 models. The agency highlighted that malicious instructions could be embedded within ordinary web content, which could lead to data leaks and unauthorized actions.

This concern reflects a broader pattern seen in nearly all major AI models released in recent years, including Google's Gemini and Meta's LLaMA. NITDA's warning emphasizes the difficulty in fully securing AI systems, which are increasingly trusted despite inherent vulnerabilities.

The article explains that AI models process material in ways that can inadvertently execute harmful instructions disguised as benign content. As AI technology evolves and is integrated into daily workflows, the risks associated with these vulnerabilities multiply.

NITDA's caution serves as a reminder that AI outputs require verification, especially when handling sensitive information, as the rapid adoption of AI in Nigeria continues.