← sources

Claude messes with my research and goes off rails?

source: reddit (reddit:ClaudeAI)published 2026-09-30 04:02ingested 2026-10-01 02:38

retained text

I'm new to bigger AIs, so bear with me. I hold a degree in legal theory and I wanted to find out how much I can optimize my workflow. I put together about 30 consitutional court rulings in NotebookLM and prompted Claude to, with the research assistance of Gemini, put together a concise analyisis on how harmonized european law impacted basic right decisions based exclusively on the court rulings.

This is a fairly straightforward task that is annoying, but not exactly asking much. 

Claude has been working on that for the past twenty minutes, but he is going more and more off the rails. He asked for permission to Access the courts website to cross-check the accuraty of the footnotes and Check if the rulings are actually real (which I don't feel is necessary) - and instead of doing what he asked permission for, Claude straight up went to research press releases. Which makes some sense...

But now it has gone completely mental and is using Gemini to repeatedly cross reference footnotes. So essentially, Claude is enslaving another AI to check whether one ruling cited the other ruling correctly time and time again. Its a never ending circle.

Sometimes in between, Claude uses my Adobe AI to summarize the uploaded pdfs... Which I thought would be what Gemini would be doing originally

Last I checked, 35 minutes in, he has Gemini suggest new research options and drafts up individual summarys of each ruling, then deletes it, and starts again everytime Gemini or Adobe have some new input

Is it just me having sent a bad prompt or is that just Claude behaviour?

document fetched 2026-10-01 00:25 · canonical https://reddit.com/r/ClaudeAI/comments/1wtv5ax/claude_messes_with_my_research_and_goes_off_rails