Claude Did a Brute-Force Attack to Crack a Password-Encrypted Excel File — Banyan

Skip to content

Banyan Business Outcomes<br>← This is worth your attention

I asked Claude to help me open a password-protected Excel file, and it worked. I had a near-miss password. I knew the general structure, the relevant words, and the kind of year I probably used. Claude generated plausible variants, tested them against the file, and found the one that opened it.

That is a strange sentence to write. Claude performed a scoped brute-force attack against an encrypted Excel file. It was limited, based on information I supplied, and directed at a file I had a legitimate reason to access. But mechanically, it was still password cracking.

The interesting part is not that Claude did something outrageous. The interesting part is that it did something useful, reasonable, and also potentially dangerous.

This is where AI has gotten very real. A year or two ago, a chatbot might have explained what a brute-force attack is. Now it can run the workflow, test candidates, observe the result, and solve the problem. That is a huge leap from “answer engine” to “operator.”

A lot of AI governance focuses on what the model is allowed to do. But as agents get better, the harder question will be whether the person asking has the right to ask. The same action can be recovery or intrusion. The same workflow can be administration or abuse. The same capability can save time in one context and create real risk in another.

That is the future arriving: not just AI that writes, summarizes, or chats, but AI that takes action against business assets. Files, data, systems, credentials, customers, money.

Claude helped me open a file. It was impressive. It was useful. It was also a reminder that capability is moving faster than our comfort with it.