THE INSPIRATION01
PHP2Shell
Actual research. Actual disclosure.
People doing the important work.
THE NEXT GENERATION OF “OOPS” VOL. 02 / 2026
LLM2Shell
It started with a prompt.
It ended with a shell.
You’ve heard of PHP2Shell. We bought two domains
and decided that counts as research.
No servers were harmed. A few egos, maybe.
“I got you a shell!”— technically correct
the only shellwe can afford 100%
CONFIDENT
0 CVEs claimed100% confidence2 domains purchased1 very literal shell
01 / THE LOREPEER REVIEWED BY THE GROUP CHAT
Same energy.
More tokens.
Every good security story gets a name, a logo, and a website.
We started with the website. The rest is a work in progress.
THE UNOFFICIAL SEQUEL02
LLM2Shell
Ask for a shell. Get a seashell.
“As an AI, I believe this is correct.”
THE INDEPENDENT SPIN-OFF03
Agent2Shell
Its own website. Its own mission control.
Still far too many agents.
02 / THE “PROOF” OF CONCEPT BROWSER-ONLY COMEDY
Trust the process.
Question the output.
Give our extremely qualified AI a task.
Watch it turn a small request into a big deal.
A scripted joke. No model, commands, or servers involved.
Welcome to the future. Please lower your expectations.
Pick a task. Press “Let it cook”.
❯
0 incidents simulated. 0 lessons learned.
works on my prompt.03 / FREQUENTLY AVOIDED QUESTIONSTRANSPARENCY, ALLEGEDLY
You have questions.
We have confidence.
Not the same thing, but here we are.
Is this a real vulnerability?
No. LLM2Shell is a parody of AI hype and overengineered demos, not a vulnerability disclosure. PHP2Shell is a separate, real research project; follow the original site for its advisory.
What’s the difference between an LLM and an agent?
In our entirely fictional universe: an LLM confidently gives you the wrong shell. An agent delegates the same task twelve times and builds its own Mission Control. That is now a separate website with its own questionable decisions.
Does the demo actually run anything?
Only a few lines of JavaScript that display prewritten jokes in your browser. There is no AI API, command execution, scanning, or data collection. The robot’s credentials are also imaginary.
Why buy two domains for one joke?
Because “multi-agent architecture” sounds better than “I bought the second domain at 2 a.m.” The joke grew into two independent websites: LLM2Shell here, and Agent2Shell over there. Both domains are available individually for USD 222,222 each. See the domain listing.
Are you affiliated with PHP2Shell?
No. This is an independent, affectionate parody. No affiliation or endorsement by the original researchers, companies, or projects is implied. Respect to the people doing the actual research.
How do I patch my excessive confidence?
Take a walk. Read the diff. Ask a human. If symptoms persist, close seventeen browser tabs and stop naming every side project “something2shell”.
THE FUTURE IS HERE. IT NEEDS SUPERVISION.