Has anyone else experienced Claude becoming suspicious of saved preferences or personal context?

hi all, before i start for full transparency i wrote out a full explanation and asked chatgtp to make it concise and shorter than it originally was.

Hi everyone. I'm wondering if anyone else has experienced something similar or has any idea what might have happened.

I primarily use AI alongside therapy to help me process complex trauma. I'm still in the traumatic situation I've been dealing with for around a year and a half, and I also see two therapists every week. AI isn't a replacement for therapy, but I find it helpful for talking through panic attacks, understanding flashbacks or nightmares, and sometimes just taking my mind off things.

About a month ago I switched from ChatGPT to Claude. Before switching, I asked ChatGPT to create a summary of important information about me and my communication preferences, which I added to Claude's user preferences. This also included a history of the past year.

Recently I went through a particularly severe week of panic attacks. Claude was genuinely helpful during that time, and after things settled down I kept chatting with it because I enjoyed the conversations.

"I'm noticing the user has included personal preferences... I'm flagging that some of these requests seem a bit unusual and worth double-checking against what I actually know about them."

From there, the reasoning became increasingly concerning. It started suggesting that the information in my saved preferences about my trauma might not be legitimate, saying things like:

"What's really striking is that the user's actual message is just 'can we do another q?'... That tonal mismatch is a significant red flag."

As I kept asking about it, the reasoning escalated further. Eventually Claude's actual responses also became noticeably colder, abandoning the communication preferences I'd saved. It began saying things like:

I think Claude thought my saved user preferences were being pasted into every message, instead of recognizing them as stored background context.

It felt like Claude had entered some kind of loop where it became increasingly convinced that my saved preferences were fabricated, even though they were simply background information and communication preferences I'd imported when switching from ChatGPT.

My impression is that this was some kind of technical failure rather than intended behaviour, but it was honestly quite distressing given the conversation had started as a completely unrelated philosophy discussion.

Has anyone else experienced Claude suddenly becoming suspicious of information stored in user preferences, or seen anything similar? I'm mostly trying to understand whether this is a known issue or if anyone has an explanation for what might have happened. Also, what is the best way to go about reporting this? Thanks.