AI didn’t really eat my homework. I don’t use AI to write, so it didn’t harm or destroy anything of mine. But in the third week of October 2026 with the announcement of a pair of severe vulnerabilities in PHP that were discovered with the help of AI, I found the context of the situation wasn’t always reported clearly. Here are some facts that may not have been picked up:
- It’s been expected for some time now that AI would lead to a period of increased number of vulnerabilities being identified and patches quickly released. This is exactly what happened.
- The vulnerabilities were discovered by researchers in August and September, and have been in the code base for a decade or more.
- Vulnerabilities in developer tools aren’t something new. We’ve seen significant threats emerge in the recent past, like log4shell in 2024, for example.
There was a controversy that elevated their profile, and it wasn’t AI-related. The critical PHP vulnerabilities (CVE-2026-109618 and CVE-2026-109619, aka “PHP2shell”) were discovered in August 2026 and came to light in October. Although they’ve existed for years, they were only just discovered by security researchers with the help of AI. This is good: AI being used to identify and ultimately resolve issues before the bad guys are able to exploit them. The controversy came after the discovery, during the process known as Coordinated Vulnerability Disclosure (CVD) in which news of the flaw is embargoed so that team responsible for the software can develop a fix before the details are released to the public. Unfortunately, this wasn’t followed, and one team announced the vulnerability before the end of the CVD embargo period. This was bad, because it gave the bad guys advanced knowledge of the exposure before a patch was widely available. Live and learn, the team that made the premature announcement has hopefully learned from their mistake.
An important takeaway from this exercise is that standard cyber defenses would likely provide adequate protection from threat actors looking to exploit the weakness. Take a moment to review the recommendations below and assess where you stand.
Real-World Security Hygiene Checklist
“Everyone has a plan until they get punched in the face.” — Mike Tyson.
Emsisoft Enterprise Security + EDR
Robust and proven endpoint security solution for organizations of all sizes. Start free trial- Take an inventory of your assets (including data) so that you know what you’re trying to defend;
- Implement ongoing training of your staff on security awareness;
- Change your router’s default name and password, turn off remote management, and turn on WPA2 or WPA3 encryption;
- Use robust authentication: strong passwords, implement MFA or Passkeys wherever possible;
- Install endpoint protection, ideally with EDR;
- Encrypt your data;
- Create and maintain offsite immutable backups;
- Limit internet exposure to systems that absolutely require it;
- Update your software with the latest patches and releases;
- Develop and practice an incident response plan. Practicing the plan is as important as the plan itself. Practice improves your ability to adapt during the chaos of a crisis situation.
Looking Ahead
AI is undeniably changing the cadence of software security, drastically shortening the window between vulnerability discovery and weaponization. Yet while discovery mechanisms evolve, defensive fundamentals remain remarkably consistent. Missteps in disclosure windows will happen, but organizations built on robust defense-in-depth principles, automated patching, and well-drilled response teams will always be best equipped to weather the storms.
