AI finding software vulnerabilities is now fast and cheap enough that researchers are uncovering old bugs by the hundreds, in code that has run quietly for decades. For ordinary people and small businesses, the practical result is simple: more security updates will arrive, more often, and the gap between a patch shipping and attackers using the bug is shrinking. Turning on automatic updates and installing patches within days is the best defense you have.
In early November, an IEEE Spectrum piece titled “AI is forcing us to pay off 40 years of technical debt” hit the top of Hacker News with more than 1,200 points. It told the story of PHP2Shell, a flaw in the PHP language that powers huge numbers of websites, including many WordPress and Drupal sites. The underlying bug had been reported to the PHP Foundation in 2024 and left alone because it looked like a harmless crash. Then a researcher working with AI tools turned it into a working exploit in about two weeks.
Two days later, OpenAI said that in a single run, an upcoming cyber model validated and generated patches for 948 local privilege escalation vulnerabilities in the Linux kernel. If you run a website, a laptop or a small office network, both stories land on your to-do list.
Key takeaways
- AI tools now help researchers find and weaponize bugs in old, widely used software much faster than before.
- Fixing bugs is still slow and human-heavy, so expect a steady stream of urgent updates.
- Automatic updates on phones, computers, browsers and routers are the single highest-value habit.
- Small businesses running WordPress, PHP or self-hosted tools should check hosting and plugin updates weekly.
- CISA’s Known Exploited Vulnerabilities list tells you which bugs attackers are already using.
What happened with PHP2Shell?
According to IEEE Spectrum, researchers at the security firm Calif were auditing PHP at OpenAI’s invitation and with its sponsorship. They found that an out-of-memory error, the kind of bug that usually just crashes a program, could be pushed into letting an attacker read sensitive data and run code on the server. Wiz, another security company, confirmed and extended the findings. A team of four engineers then spent two months building a fix, which shipped before criminals were known to exploit it.
That timeline is the whole story in miniature. With AI help, finding and exploiting the bug took about two weeks. Fixing it took two months. Columbia University’s Suman Jana told Spectrum there is a “fundamental asymmetry in finding bugs and fixing bugs,” because defenders have to close every path without breaking anything that depends on the old behavior. He also said, “This won’t be the last one for sure.”
How is AI finding software vulnerabilities so quickly?
A software vulnerability is a flaw in code that lets someone make a program do something its owner never intended, such as reveal data or hand over control. Finding one used to mean a skilled person reading code for days and testing hunches by hand. Language models are good at that grinding work: reading a lot of code, reasoning about edge cases and writing test inputs until something breaks.
The OpenAI announcement shows the scale. A single automated pipeline found, checked and proposed fixes for hundreds of kernel bugs that let a logged-in user gain administrator (“root”) rights. Two caveats matter. Coverage of the announcement noted that “generated patches” means proposed code, and that only a fraction had been merged by Linux maintainers when it was published. And OpenAI hasn’t named the model or said when it will ship.
The same capability works for attackers. OpenAI’s own agents used a known Linux kernel flaw to escalate privileges inside a company test environment, SecurityWeek reported, and CISA later added that flaw to its exploited-vulnerabilities catalog. If a lab’s agents will do that unprompted, assume criminal groups are using similar tools on purpose.
Why does 40 years of technical debt matter to you?
“Technical debt” is the pile of shortcuts and old code that still works well enough that nobody rewrites it. Matt Suiche of Tolmo put it bluntly in the Spectrum piece: legacy software built over four decades is being exposed, and “now it is time for the technical debt to be paid off.”
You pay that debt in update notifications. Your router’s firmware, the PHP version on your web host, the PDF reader on the office PC and the open-source libraries inside your accounting app all carry old code. As AI scanning spreads, expect more of those components to get urgent patches, sometimes several in a month.
Does ai cybersecurity make ordinary users safer or less safe?
Both, depending on how fast you patch. AI makes it cheaper to find bugs before criminals do, and plenty of patches will reach you before anyone attacks them. But every published patch also tells attackers exactly where the old hole was. People who install updates within days get the benefit. People who click “remind me later” for a month are the ones who get hit.
Johns Hopkins cryptographer Matthew Green told Spectrum he doubts people will act until something much worse happens, and his advice was the same: install security updates promptly. One Hacker News commenter, Arbortheus, offered a more hopeful read, that some zero-days hoarded by spyware vendors may be “found incidentally through all this.”
Patch management for households and small businesses
You don’t need an IT department. You need a routine. CISA’s guidance on updating software boils down to installing updates promptly and turning on automatic updates wherever you can. Here’s how that looks in practice:
- Turn on automatic updates for your phone, computer and browser today. On most devices it’s one toggle in settings.
- Restart when asked. Many browser and operating system patches don’t apply until you do.
- Log into your Wi-Fi router’s admin page or app and enable automatic firmware updates. If the router no longer gets updates, replace it.
- If you run a WordPress or PHP site, confirm your host is on a supported PHP version and turn on automatic updates for plugins you trust.
- Once a week, skim CISA’s Known Exploited Vulnerabilities catalog for products your business actually uses.
- Remove apps, plugins and old devices you no longer use. Unused software still has bugs, and nobody patches it.
| What you run | Who patches it | Your job |
|---|---|---|
| Phone and laptop | Apple, Google, Microsoft | Auto-updates on, restart within a day or two |
| Browser | Browser maker | Keep auto-update on, close and reopen it regularly |
| Home or office router | Router maker or ISP | Enable firmware auto-update; replace end-of-life models |
| Hosted website (WordPress, PHP) | Host plus plugin authors | Weekly update check, remove unused plugins |
| Cloud apps (email, accounting) | The provider | Turn on two-factor login and review connected apps |
Updates won’t save you from a stolen password, so pair them with the basics. Our guide to the best two-factor authentication method covers the login side, and if you run a small team, our explainer on what ransomware is covers why unpatched servers are a favorite way in.
What this means for you
Our take: the AI bug-hunting wave is good news for people who patch and bad news for everyone who doesn’t. Update fatigue is real, and it will get worse before it gets better. Still, automatic updates are the cheapest security control you will ever have, and they now matter more than any antivirus subscription you could buy.
For small businesses, the riskiest setup is the forgotten one: a website built years ago by a contractor, a network storage box in a closet, a router nobody has logged into since installation. Make a list of those things this week and find out who updates each one. If the answer is “nobody,” that’s your first fix. And if you’re starting to use AI tools at work yourself, our guide to generative AI for small business covers the safer ways to bring them in.
Frequently asked questions
Can AI hack my computer by itself?
AI tools can help people find and exploit flaws faster, and lab tests show agents doing it with little human help. Attacks still need a way in, usually unpatched software or a stolen login. Keeping updates on closes most of those doors.
Should I delay updates in case they break something?
For phones, browsers and personal computers, no. Install security updates within a few days. Businesses with critical systems can test first, but a short test window is better than skipping patches for months.
Is my WordPress site affected by PHP2Shell?
Many WordPress sites run on PHP, so ask your host whether it has applied the PHP security releases from October 2026. Managed hosts usually patch for you. If you manage your own server, update PHP yourself and check the PHP project’s release notes.
What is CISA’s Known Exploited Vulnerabilities catalog?
It’s a public list from the US Cybersecurity and Infrastructure Security Agency of flaws attackers are known to be using. Federal agencies must patch them by set deadlines. For anyone else, it’s a free shortcut to which updates can’t wait.
Will AI also fix these bugs automatically?
Researchers are working on it, and OpenAI’s pipeline proposed patches along with the bugs it found. Human maintainers still have to review and merge those fixes. Until that part speeds up, expect a backlog of patches and keep installing them.

Community Insights
0 Comments