"I Heard AI Agents Are Hacking the Banks." Is That True? The PHP2Shell News, Explained Simply

Maybe you saw a headline. Maybe a friend said it out loud: "I heard AI agents are hacking the banks. Is that true? What does that mean?"

It is a fair question. This week had a lot of scary computer news, and it all got mixed together. This page sorts it out in plain words. It is meant to be easy to read for anyone, including kids and parents reading together. If you want the full technical version, we have one: the PHP2Shell guide.

Is It True That AI Agents Are Hacking the Banks?

Not as far as we can find. None of the news reports we read said that banks were broken into because of this week's problem. None said that AI agents were doing the attacking.

Here is what the reports did say. Two serious flaws were found in a piece of software called PHP. The people who make PHP fixed them on Oct. 21, 2026. Some attackers tried to use the flaws right away. And AI helped the researchers who found the flaws.

So there is a true story under the headline. It is just a different story.

What Is PHP?

Think of a website like a restaurant. You ask for something, and the kitchen makes it and sends it out.

PHP is a tool that works in the kitchen. When you open a page, PHP is often the one that puts the page together before it is sent to you. About 70 out of every 100 websites whose kitchen tool is known use PHP. That includes WordPress, a very popular way to make websites.

Because PHP is in so many kitchens, a problem in PHP is a problem for a lot of websites at once.

What Is a "Bug," and What Does "Hacking" Mean?

A bug is a mistake in a computer program. Most bugs are small. Some are big.

Picture your front door. It has a lock. Now say the lock has a tiny mistake in how it was built, and a stranger who knows the trick can open it without a key. That is a bad bug. It is a flaw in the lock itself, not something you did wrong.

Hacking is when someone uses a flaw like that to get into a computer system they are not allowed to use.

The two flaws this week are called PHP2Shell. The people at PHP rated them "critical," which is the most serious label. They said an attacker might be able to tell a website's computer what to do without ever logging in. The experts call that remote code execution. It means someone far away gives orders to a computer that should not listen to them.

Where Does AI Come In?

This is the part that got twisted in the headlines.

A security company named Wiz said it used AI to help find the flaws. It said it used its own tools plus two AI models, Google's Gemini 4 Argon and OpenAI's GPT-5.6 Sol. Wiz also said AI makes it easier to turn old, hidden flaws into working attacks.

OpenAI told the news site Axios: "To our knowledge, frontier AI found these vulnerabilities before attackers did."

Put simply: AI helped the good guys find the broken lock first. That is not the same as AI robots breaking into banks.

Could attackers use AI too? Wiz thinks AI is speeding up the work, both for finding flaws and for using them. But we do not know how many attackers are using AI, and the reports we read do not say that AI agents carried out these attacks.

Were Any Attacks Real?

Yes, some tries were real. The reports say:

• On Oct. 17, a trap set by Amazon recorded one try to use the flaw before anyone knew about it publicly. Amazon described it as mostly a one-off test.
• On Oct. 21, the day of the fix, Cloudflare reported early attacks. Cloudflare is a company that helps protect websites.
• A security company named Patchstack said hosting companies reported real attacks. It said its own data only showed people poking around at that time.
• Axios reported on Oct. 22 that the real harm has been limited so far, and that many organizations have already reduced the risk.

We do not know how many websites were broken into. Nobody has shared a full count.

So Are Banks Safe?

We cannot say for any one bank. Here is what we can say.

A bank is a business, and businesses use many computer systems. The news reports we read did not name a bank as a victim of PHP2Shell. Many big organizations have security teams whose job is to install fixes quickly. Still, a fix only helps after it is installed, and some older servers around the world do not get fixes at all.

If a company you use finds out that your information was affected, it will usually send you a notice. Read it when it comes. Do not ignore it.

What Can You Do?

You cannot fix a company's computers. But you can make it much harder for a stranger to use your accounts. These steps are easy:

1. Use a different password for every account. Then one stolen password cannot open the rest. A password manager can remember them for you, so you do not have to.
2. Turn on two-step sign-in. This is also called multi-factor authentication. It asks for a code or a fingerprint in addition to your password. Do it first for your email and your bank. Passkeys work well too.
3. Be careful with messages about "the big hack." Scammers love big news. If an email or text says your account was hacked and tells you to click a link, do not click. Go to the company's website yourself or call the number on your card.
4. Change your password if a company tells you to. Read the message first, and make sure it is real by going to the site yourself.
5. Stay calm. The experts Axios talked to say not to panic. The harm so far has been limited.

What If You Run a Website?

If you or someone you know has a website, ask the person who runs the server one question: "Did we update PHP?"

The fixed versions are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35, or a fixed package from your server company. Updating WordPress or a plugin does not fix it. The fix has to happen on the server, and the server must be restarted so the new version starts. Our full guide has a longer checklist. A website with no PHP at all, such as a simple page made of plain files, is not affected by these flaws.

Words to Know

• Bug: a mistake in a computer program.
• Vulnerability: a bug that an attacker could use to cause harm.
• Hacking: using a flaw to get into a system without permission.
• Patch: a fix for a bug. "Patching" means installing the fix.
• Server: the computer that holds a website and sends it to you.
• AI agent: an AI program that can take steps on its own, such as running tools, instead of only answering questions.

What We Can and Can't Say

This section is analysis, not reporting.

• The headline and the facts do not match. "AI hacking the banks" sounds like a movie. The facts we found are about a software flaw, a quick fix, some real attack attempts and AI that helped find the problem.
• AI changed the speed, not the basic rules. Finding and fixing flaws is old work. The new part is that AI may let both defenders and attackers do it faster. That is a reason to install fixes quickly.
• We do not know everything. We do not know how many sites were hit, who the attackers were, or whether any bank was affected. If those facts come out, we will update this page.

Why It Matters

• Headlines can mix up several stories. This week's AI, PHP and bank talk are three different things.
• A flaw in a common tool touches a lot of websites. PHP is in about 70 out of 100 sites whose kitchen tool is known.
• AI is on both sides. It helped find the flaws first, and it may speed up attackers too.
• Simple habits protect you. Unique passwords and two-step sign-in help no matter what the headline says.