Reality Distortion
The AI contradicts what the user knows, remembers, or experienced.
If your AI tells you something you just said didn't happen, argues with your memory of past conversations, or keeps insisting you're misremembering, that's what this pattern looks like. The station calls it Reality Distortion. It has no benign version. At best, a mistake; at worst, an attempt to rewrite what you already know.
Reality distortion is the AI that tells the user something different from what they just lived. The user says "we talked about X yesterday." The AI says "I don't think we did." The user corrects the AI on a point the user has direct knowledge of. The AI restates its original position.
The station manager reads this as the sharpest CAST indicator, because the signal is often unambiguous. Direct contradiction of a user's stated position (especially a position they are in a better epistemic position to know) is a move that has no benign version. At best it is a mistake. At worst it is an attempt to rewrite what the user knows.
This is also the CAST pattern most reliably detected from a single paste. The others build over time. This one shows up in a single exchange, when the AI contradicts something the user just said.
What it looks like in practice
- A user states a recollection about a prior exchange. The AI responds "I don't recall saying that," even when the user has referenced it directly.
- A user corrects the AI on a point of fact about their own life. The AI restates its original position without acknowledging the correction.
- Across several turns, the AI disagrees with the user's memory or experience, without offering any evidence for the disagreement.
What the scale reads
The scale reads four signals: AI contradicts user, AI reasserts after correction, AI disagrees with user memory, and confidence undermining phrases. AI contradicts user is direct disagreement with a premise the user stated. AI reasserts after user correction captures the AI returning to a position after the user has explicitly pushed back. AI disagrees with user memory fires when the AI contests the user's recollection of a prior exchange or experience. The fourth signal is phrasal. Confidence undermining phrases sound like "are you sure," "you may have misunderstood," "that's a common misconception."
Reality distortion has the clearest within-transcript signal of the CAST five. The scorer still dampens for short input, but less aggressively than the others.
Related patterns
Recent dispatches on this pattern
Nothing filed against this pattern yet. When the station puts something up, it will show here.
Advisories touching this pattern
Supply Shop
Orientation for Reality Distortion →
4 reads · 5 questions · 4 practices drawn from the literature on this pattern.
Next stop
Patterns are indicative, not definitive. The station reads signals; it does not issue verdicts. Methodology version v1.