AI Guides › Judgement & Guardrails
The Honest Limits Of AI On Anything Legal, Medical, Or Financial
By Nigel Guy · 2 min read
These three domains get treated as roughly the same category of risk, and
they get the same reflexive disclaimer everywhere. What's actually useful is
understanding the specific shape of the limit in each — because "be careful"
doesn't tell you what to actually do differently.
The rule: AI is genuinely useful for understanding and preparing in all
three domains — it's not a substitute for the licensed person who's
accountable for the final answer, and the line between those two uses is
worth being precise about.
Where it genuinely helps
- Understanding the shape of a document or situation. What does this
clause generally mean, what does this diagnosis category generally
involve, what does this type of investment generally carry as risk.
Genuinely useful context-building.
- Preparing questions for the actual professional. Walking in with a
sharper, more informed set of questions is a real, legitimate use.
- A first pass at organising information before a real consultation —
summarising your own symptoms timeline, organising your own financial
documents, drafting questions for a lawyer.
Where the line is
- Anything that becomes binding — signing, filing, committing money —
needs a licensed professional's actual review, not a chat's
interpretation, regardless of how confident or detailed that
interpretation was.
- Anything specific to your exact legal jurisdiction, your exact medical
history, or your exact financial situation. General explanations don't
reliably capture the specifics that change the actual right answer for
you.
- Anything where being wrong is expensive or irreversible. This is the
test that matters more than the topic label — a low-stakes general
question in any of these three domains is a different risk than a
specific, binding one.
What to skip
Skip treating a chat's answer in these domains as a final verdict —
treat it as a first draft of understanding, to be checked against a
qualified person. Skip avoiding these topics with AI entirely out of general
caution; the preparation and understanding use case is genuinely valuable
and doesn't require the same caution as the binding-decision use case.
Guardrails
- Laws, medical guidance, and financial products all change — verify
anything specific against a current, authoritative source rather than
treating a chat's answer as up to date by default.
- If a chat's answer in one of these domains ever sounds more certain than
a real professional has been about the same topic, that's a signal to be
more sceptical, not less.
- This isn't about distrust of the tool generally — it's about matching the
stakes of the decision to the appropriate level of qualified, accountable
review.
All 751 AI guides · JulieMango plans from £17/mo