On July 18, 2026, Kenyans who visited president.go.ke did not find the Head of State. They found a ransom note. The official State House homepage had been defaced and replaced with demeaning messages targeting President William Ruto, a cryptocurrency wallet address, and a demand for five bitcoin, roughly Sh41 million.

State House confirmed it was aware of the breach and said its ICT team was working to restore the site.

It was not the first time. In November 2025, president.go.ke went down alongside the ministries of Interior, Health, Education and ICT, the Directorate of Criminal Investigations and the Immigration Department, in a coordinated defacement that scrawled neo-Nazi slogans across government pages.

Two breaches of the presidency's own digital front door inside a year is not bad luck. It is a pattern, and the pattern is the story. Here is the uncomfortable part. On the evidence available so far, the State House defacement looks like a conventional attack: exposed surfaces, weak credentials, opportunistic extortion. No one has claimed it was powered by artificial intelligence, and no responsible analyst should pretend otherwise. But that is precisely why it matters.

The techniques that felled the presidential website are the exact fundamentals that AI is now learning to exploit at industrial scale and machine speed. If the old methods can reach the President, the new ones will reach almost anyone.

Consider the evidence from the past nine months. On 21 February 2026, CJ Moses, Chief Information Security Officer of Amazon Integrated Security, published a detailed threat intelligence report documenting a five-week campaign in which a single Russian-speaking threat actor compromised more than 600 Fortinet FortiGate firewalls across 55 countries.

The attacker exploited no zero-day vulnerability and deployed no novel malware. According to Amazon's assessment, the individual possessed only "low-to-medium" baseline technical capability, significantly augmented by artificial intelligence. They scanned exposed management interfaces, brute-forced weak credentials that lacked multi-factor authentication, and used a custom Model Context Protocol server to feed reconnaissance data into commercial large language models, which generated structured attack plans for domain compromise, lateral movement and data exfiltration.

The attack's sophistication was not in the technique. It was in the tempo. In five weeks, a single actor achieved what would previously have required a professional team operating over months. The AI-generated tooling was, by conventional standards, mediocre. Amazon's analysts noted redundant comments and naive code. But mediocre code deployed at machine speed across 55 countries is not mediocre. It is a paradigm shift.

These two incidents sit at opposite ends of the capability spectrum, a lone amateur and a state-sponsored group, but they deliver the same verdict. AI has collapsed the barrier between intent and capability in offensive cyber operations. The skill premium that once separated a curious teenager from an advanced persistent threat is being eroded by tools available to anyone with an internet connection.

The exposure is not limited to attackers wielding AI. The AI agents themselves are becoming the target. OpenClaw, an open-source autonomous assistant that crossed 180,000 GitHub stars in January 2026, was found within weeks to carry a critical vulnerability enabling one-click remote code execution.

By early February, more than 40,000 instances were exposed online, 63 per cent of them assessed as vulnerable. A security audit of its plugin marketplace found that 12 per cent of listed skills were malicious. By April, the tool had accumulated 138 disclosed vulnerabilities. As AI agents gain the ability to browse, execute code, manage files and act on a user's behalf, the attack surface is no longer the application. It is the entire digital life of the person, or the institution, that installed it.

Nairobi. The frameworks that would have protected the presidential website are neither exotic nor expensive. The OWASP LLM Top 10, the NIST AI Risk Management Framework and joint guidance from the United Kingdom's National Cyber Security Centre and the United States' Cybersecurity and Infrastructure Security Agency all converge on the same unglamorous fundamentals: multi-factor authentication, least-privilege access, continuous monitoring, prompt patching, and treating every AI agent as untrusted code. None of this is new. All of it was available before Saturday morning.

The cybersecurity doctrine of the pre-AI era assumed that skill was the bottleneck. It no longer is. When a defacement can reach the President twice in one year, and when a low-skill actor abroad can compromise 600 firewalls in five weeks, the bottleneck is not talent. It is hygiene. Exposed interfaces, single-factor logins, unpatched systems: these are not sophisticated failures. They are the fundamentals, and AI has turned neglecting them into an existential risk.

For governments, banks, hospitals and businesses across Africa, the lesson from president.go.ke is not that the enemy is unusually clever. It is that the enemy no longer needs to be. The barrier to entry has not merely lowered. It has, for practical purposes, been removed. The only question that matters now is whether our defences are ready for a world in which everyone can attack.

-The writer is founder of Concept Vault Limited.