I would like to discuss a change I have noticed in recent ChatGPT versions, especially after GPT-5.1, when using it for legal drafting and administrative correspondence.
In complex legal conversations, there are often two very different layers:
-
Internal strategy discussion with the user:
- procedural risks,
- tactical considerations,
- possible weaknesses,
- how to protect the user’s rights,
- what should not be revealed to the opposing party.
-
Final text intended to be sent externally:
- to a lawyer,
- to an administration,
- to a court,
- or to another official body.
The problem I have noticed is that recent versions sometimes blur these two layers. The model may insert internal strategic reasoning, emotional framing, or tactical concerns into a final letter or legal draft. This can be risky, because a final legal or administrative letter should be clean, professional, focused, and should not reveal the user’s internal strategy.
I also noticed another issue: legal drafting sometimes became too neutral or cautious. Instead of using stronger and more protective legal wording such as “violation”, “breach”, “failure to comply with”, “infringement of the right of access to a court”, or “denial of an effective remedy”, the model often uses softer expressions such as “error”, “difficulty”, or “issue”. In legal pleadings, this can weaken the user’s position.
For legal use cases, the model should be better at distinguishing between:
- analysis for the user,
- risk assessment,
- drafting strategy,
- and the final document to be sent externally.
A useful improvement would be a clearer drafting mode where ChatGPT:
- keeps internal strategy out of the final text,
- flags procedural risks clearly,
- uses precise and assertive legal language when appropriate,
- and produces final letters or pleadings that are protective of the user’s rights without being exaggerated or emotional.
Has anyone else noticed this change in recent versions?