Claim
“The Federal Interface Commission was founded in 2017.”
Public-interest software proposal
AI hallucination exit system
When an answer invents, show people how to get out.
Hallucination Trace is a proposed verification layer that explains why an AI claim may be unreliable—before a user gets caught in a rabbit hole.
A visible correction record
One claim.
One clear exit.
The software does not simply flash a warning. It shows the user what may have gone wrong and what to do next.
Trace 0001Evidence gap detected
Status
Unsupported
Likely mechanism
Entity collision
What happened
The answer appears to combine the name of one organization with the history of another.
Correction
No authoritative record supports this claim. Remove it or verify it with a primary source.
Next action Search primary sources, revise the claim or remove it.
Three moves
Interrupt the spiral.
Turn an uncertain answer into a short, inspectable path back to evidence.
Isolate
Separate factual claims from interpretation, speculation and creative language.
Check
Compare each claim with trusted evidence and test whether the answer remains stable.
Explain
Name the likely failure pattern, show the evidence gap and provide a correction route.
What it can reveal
Evidence, instability, likely failure pattern, uncertainty and correction.
What it cannot claim
Direct access to a model’s private reasoning or absolute certainty about why a token appeared.
The public promise
Do not ask people to trust the warning. Show them the trail.
Designed for journalists, students, researchers and anyone who needs to know when fluent language has outrun the evidence.