A Skeptic Put Up $500 in Bitcoin as Bait—and Proved Himself Wrong: PHP2Shell Really Can Be Exploited

A couple of weeks ago, two critical PHP vulnerabilities prompted a significant number of websites to go offline for updates. Immediately after the announcement, “experts” appeared claiming that the situation was not so serious and that exploiting the flaws in the real world was nearly impossible. Their main argument was that nobody had published a working exploit. One skeptic decided to put that claim to the test.

The $500 Bait

The author built a website, placed the private key to a Bitcoin wallet containing $500 on it, and posted the link on Hacker News: hack it and take the money. Nobody succeeded during the first few hours, and commenters naturally saw that as confirmation that they had been right.

But the reasons were mundane. First, the site was flooded with traffic, kept crashing, and left many requests unanswered. Second, the site itself was not vulnerable. PHP was installed and even served the homepage, but its configuration did not match the one affected by the vulnerability.

The author gave the server more resources and rewrote the code, which had been too simple and unlike typical websites. After that, the vulnerability was successfully triggered and a hacker obtained the key. Admittedly, the author had ultimately provided the wrong key, but that is beside the point.

The Skeptic’s Admission

In his final blog post, he wrote that his view had been wrong: PHP2Shell remains a real threat and can be exploited. He also believes exploitation is still difficult and that a person without expertise, armed only with an LLM, would most likely fail.

That conclusion is debatable. Most attempts were probably not serious: people did not spend many tokens or use the most capable models. Moreover, such models often block hacking requests.

There is data as well. An expert who hacked OpenAI several months ago reports that, in his tests, only Mythos and Sol-5.6 can execute the attack within 2–4 hours. The open GLM-5.3 and Kimi K3.1 models ran for more than 20 hours and failed, and he doubts that open models are capable of the task at all.

This also answers a popular question: why are there no reports of hacks coming from China—are Americans merely hyping the issue? Judging by these tests, hype is not the explanation: the models available there still lack the capabilities required for this task.

What to Do in Practice

First, do not wait for a public exploit before updating. Video demonstrations of attacks against real products in the original announcement are already sufficient evidence. The absence of code means only that the code has not been published.

Second, if you are testing someone else’s vulnerability claims, reproduce the production configuration rather than a simplified test environment. The presence of PHP on a server does not by itself mean that the server is affected. That was exactly the skeptic’s mistake, and defenders who assume they are safe can make the same one.

Third, assess risk with the capabilities of frontier models in mind. A few hours of exploitation time on a powerful model is inexpensive compared with the cost of downtime. Weak tests using cheap models and a small token budget create a false sense of security.

Limitations

This is the story of one experiment and one set of tests. The model data came from a single researcher, not an independent benchmark. Exploitation still requires expertise and resources. But the conclusion is simple: update before waiting for an elegant proof.