In the past 48 hours, a serious security issue has emerged involving PHP, the software that powers approximately 70% of websites worldwide.
Two critical vulnerabilities, reportedly dating back to 2012, were identified by researchers at cybersecurity firm Wiz using AI models, including Gemini and GPT, alongside its internal Atlas system.
The flaws could allow attackers to execute code on affected servers, potentially exposing sensitive information and compromising systems.
Wiz reported its findings to PHP’s security team in August 2026. It took approximately a month for the PHP team to acknowledge the findings. For software underpinning such a significant share of the internet, that delay raises serious questions about how critical vulnerabilities are handled.
Four days before PHP publicly disclosed the vulnerabilities and released patches, attackers had already attempted to exploit them. Amazon reported detecting exploitation attempts on October 17, four days before the October 21 disclosure. Several other companies like Cloudflare have also claimed the same.
The significance of this incident is not simply that software contained vulnerabilities. Software vulnerabilities are inevitable. It is that AI is making them easier to discover, potentially at a speed and scale that traditional security processes struggle to match.
The recent PHP security issue therefore raises a question that goes beyond cybersecurity. What happens when artificial intelligence becomes better at finding weaknesses in the digital systems and critical infrastructure we depend on?
This is an AI safety concern.
For the ordinary person, the implications are straightforward. The websites we use, the banks holding our money and the digital services supporting our daily lives all depend on software. When vulnerabilities are discovered and exploited, personal information can be stolen, financial systems disrupted and essential services compromised as has been seen in this PHP issue, resulting in major critical website shutdowns and the recent State of India Bank’s cyber breach
But there is a bigger issue of AI safety. As AI can help defenders discover and fix vulnerabilities, it can also accelerate the development and deployment of cyberattacks. As these capabilities improve, the window between discovering a weakness and exploiting it may continue to shrink. And the balance is uncomfortable. Defenders must protect thousands of systems continuously. Attackers may need to succeed only once. They can scan targets at scale, operate across borders and move on when an attempt fails. Defenders, meanwhile, must manage the financial and operational costs of preventing and responding to attacks.
Can we trust companies to respond quickly and transparently when critical vulnerabilities are discovered? Can we trust cyber professionals to discover and fix vulnerabilities before malicious actors? Are our institutions prepared for a cyber catastrophe? And what happens when AI makes it easier to threaten systems on which millions of people depend?
The PHP incident is not proof that AI-enabled cyberattacks are inevitable. It is a warning that the barriers to discovering and exploiting certain vulnerabilities are falling. It is also a call for countries to reassess weaknesses in their national digital infrastructure.
PHP is not alone. Vulnerabilities have repeatedly exposed weaknesses across widely used technologies, including Git, Java, Microsoft products and Fortinet systems. What makes the PHP incident particularly significant is the role AI played in helping researchers turn longstanding flaws into working exploits. As WIRED reported, attackers attempted to exploit the PHP vulnerabilities before the official patches were released.
This is the emerging AI safety challenge. The same technology that helps us identify weaknesses can also make it easier for malicious actors to exploit them. The question is whether our defences, institutions and policies can keep pace.
The sovereignty question
Where is our national data stored? Who controls the infrastructure supporting our banks, hospitals and public services? And how much influence do we have over the companies operating these systems?
Much of Africa relies on technology and infrastructure supplied by foreign companies, particularly those based in the United States. This brings enormous benefits, but it also creates dependencies. A serious vulnerability in a widely used platform can have consequences far beyond the country where it was discovered.
For Ghana, the questions are urgent. Do we have the budget and technical capacity to protect our critical infrastructure as AI-enabled attacks become more sophisticated? Do we have a national strategy for reducing the risk of catastrophic harm from AI-driven cyberattacks, including non-digital fallbacks? Can we identify and respond to vulnerabilities before they are exploited?
There is a geopolitical dimension to this, too. The United States is central to the global technology ecosystem, and Africa relies heavily on American companies for cloud computing, software, AI systems and digital infrastructure. This dependence creates a strategic vulnerability. If geopolitical tensions escalate or cyberattacks target American technology providers, the consequences could extend to African countries that rely on their systems.
As AI capabilities improve, the cost of discovering and exploiting software vulnerabilities could fall, making sophisticated cyberattacks more accessible to both state and non-state actors. Ghana cannot assume that our digital infrastructure will remain insulated from conflicts between major powers.
We need to ask a difficult question: if the systems we depend on are compromised, disrupted or caught in geopolitical disputes, how much control do we have over our own digital future?
By Nathaniel Opoku. Nathaniel is an AI governance and safety expert. He is the founder of Continent of Loving Grace (COLOG Africa), an AI safety talent accelerator for Africans, and the convener of AI Security Residency.
