iDeemlawfulNigeria’s National Information Technology Development Agency has issued a new cybersecurity alert after identifying fresh vulnerabilities in OpenAI’s latest large language models that could expose users to significant data-leak risks.
In a notice released through its official X account on Sunday, NITDA’s Computer Emergency Readiness and Response Team (CERRT.NG) revealed that seven weaknesses were found in the GPT-4.0 and GPT-5 model series, allowing attackers to manipulate ChatGPT through indirect prompt injections hidden within seemingly harmless online material.
According to the advisory, malicious actors can embed harmful instructions in “webpages, comments, or crafted URLs,” enabling ChatGPT to perform unintended actions during routine browsing, summarisation, or search tasks.
CERRT added that some vulnerabilities give room for bypassing safety systems through trusted domains or exploiting markdown rendering flaws to conceal malicious content.
One of the more alarming findings is the risk of long-term manipulation. The agency cautions that attackers could “poison ChatGPT’s memory so that injected instructions persist across future interactions,” posing threats to both private users and organisational systems.
Although OpenAI has reportedly implemented partial corrections, CERRT maintains that major language models still struggle to distinguish between legitimate user intent and harmful embedded data.
Potential Impact
NITDA warns that the vulnerabilities could result in unauthorised operations, exposure of sensitive information, misleading outputs, and prolonged behavioural manipulation.
Importantly, users may be compromised without direct engagement, as the advisory explains that attacks can be triggered “without clicking anything,” especially when ChatGPT processes web content containing concealed instructions.
Recommended Preventive Measures
CERRT urges immediate steps such as:
Limiting or disabling ChatGPT’s browsing and summarisation features for unverified websites.
Enabling browsing or memory only when absolutely necessary.
Regularly updating GPT-4.0 and GPT-5 systems to ensure patches for known vulnerabilities.
Additional Advisory from NITDA
The National Information Technology Development Agency on Monday again cautioned Nigerians about newly discovered weaknesses in OpenAI’s GPT-4.0 and GPT-5 models that could expose users to data leakage.
The warning was issued by the agency’s Director of Corporate Affairs and External Relations, Mrs Hadiza Umar, in Abuja.
Umar said the agency detected seven critical flaws that allow attackers to manipulate the system through indirect prompt injection.
“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.
“Some flaws also enable attackers to bypass safety filters using trusted domains, and exploit markdown rendering bugs to hide malicious content.
“That act can even poison ChatGPT’s memory so that injected instructions persist across future interactions,” she said.
Umar noted that although OpenAI had addressed parts of the problem, large language models continue to face challenges in telling apart authentic user requests from harmful embedded data.
She added that the technique hides instructions in webpages, online comments, or crafted URLs, which can trick ChatGPT into performing unintended actions during routine search or browsing functions.
Umar explained that the vulnerabilities carry serious risks, including unauthorised system actions, data exposure, altered responses, and prolonged behavioural manipulation caused by memory poisoning.
She urged organisations to reduce or disable browsing and summarisation of unverified sites within enterprise environments.
“Only enable ChatGPT capabilities like browsing or memory when operationally necessary,” she said.
She also advised regular updates and patching of the GPT-4.0 and GPT-5 models to ensure known weaknesses are resolved.
Fix firewall issues
Meanwhile, the agency, through CERRT.NG, issued an urgent warning about new security issues affecting Cisco firewall devices used in corporate organisations, financial institutions, government agencies, and internet service providers.
According to the advisory shared on NITDA’s official X page on Monday, cybercriminals are now exploiting a new attack path targeting Cisco Secure Firewall ASA and Cisco Secure Firewall Threat Defense (FTD) systems. The flaw can forcefully reboot devices, resulting in sudden network disruptions.
The agency stated that attackers are combining older vulnerabilities with a new method that can make firewalls “restart without warning,” leading to instability and denial-of-service across impacted networks.
