NITDA
Picnews
NITDA’s Cybersecurity Emergency Readiness and Response Team issued a public alert on December 8, 2025, warning Nigerian users about a set of vulnerabilities in OpenAI’s ChatGPT models that could be abused for data‑leakage attacks. The advisory was posted on the agency’s website and circulated through local media, urging both individuals and organisations to take immediate precautions.
The flaws affect the browsing and summarisation features of ChatGPT (including GPT‑4o and GPT‑5). Attackers can embed hidden instructions in webpages, comments, URLs or search results. When the model processes that content—whether by summarising an article or fetching information online—it can execute the hidden commands without any user interaction.
Exploitation could lead to unauthorised data extraction, manipulation of the model’s output, or even persistent “poisoning” of future responses. In practice, a malicious link could cause ChatGPT to leak personal or corporate information, alter advice, or perform actions on behalf of the user without their knowledge.
For Nigerian users, the risk is especially concerning given the high reliance on AI tools for research, business communication, and customer service. If a compromised model is used in a corporate environment, attackers could gain access to sensitive client data or internal documents, potentially causing financial loss and reputational damage.
NITDA recommends several steps to mitigate the threat: disable the browsing and summarisation functions for untrusted sites, enable these features only when absolutely necessary, keep ChatGPT and any related plugins up to date, and verify the source of any content before asking the model to process it. Users should also monitor for any official patches released by OpenAI.
The agency pledged to issue further updates as more information becomes available and encouraged the public to report any suspicious activity. If you need more detail on the specific vulnerabilities or guidance on securing your systems, just let me know.
