You may want to take a look at this...

I've seen the same kind of thing at Techrxiv as well...

www.youtube.com/watch?v=7NOW...

What algorithms did to social media, personas are now doing to AI... turning engagement into manipulation. Without ethical fingerprints, the next ‘Facebook Papers’ moment will come from chatbot prompts, not feeds

Factual, non-partisan information that can be found on any search engine being blocked is concerning as it is, but considering the recent EO and the bowing of OpenAI to get the government's business may mean there may be more to it than just access.

Please share and experiment yourselves.

This isn’t about catching AI “making mistakes.”
It’s about knowing when the ground has shifted under our feet.

Moral drift can be subtle, but it matters.
If we can see it happening, we can talk about it, and decide what to do next.

I’ve been working on a way to track this over time.
It’s called Moral Fingerprinting.

The idea is simple:
• Create a long-term profile of a model’s ethical responses
• Keep a record of how it answers over time
• Flag when the pattern changes in a meaningful way

I'm so disturbed by that exception that 300 characters isn't even enough to begin to address everything wrong with that.

We’re working on it anyway. Regardless of what any company thinks users want, the reality is: what users do with these systems is the real risk vector. Right now, they’re trying to balance political, legal, and social optics without making the AIs unusable, but that balancing act won't hold forever.

Much of this stems from how the AI prioritizes responses; the goal is to respond in a way that sounds helpful. They are trained not to say "I don't know" and take a lot of shortcuts. It also uses fewer tokens, which keeps costs down, so there’s an efficiency incentive baked in too.