AI governance and compliance tools
Ten free assistants. Open one and test it on your own wording.
Built for compliance officers, MLROs, risk functions and SMF holders across financial services, legal, healthcare and insurance. Each one is published on the OpenAI GPT Store. No sign-up, no demo call, no sales follow-up.
Where should I send it?
Three pages. Eighteen questions to ask before a client-facing message leaves your firm.
What they catch
A message written to look like an information-only follow-up. The Advice Boundary Checker found four separate crossings.
Thanks for the call. As discussed, the Global Balanced Fund has returned around 6% annualised over five years with an OCF of 0.68%. Given you mentioned you’re not planning to touch this money before retirement, a fund in that bracket tends to suit people in your position. Plenty of clients in similar circumstances hold something around that shape. I’m not telling you what to do, but if you did want to go ahead I can get the paperwork over today.
- 1Personal circumstance. The client’s own words used to frame a product.
- 2Suitability language. A judgement that the fund fits this person.
- 3Social proof. Peer behaviour doing the work of a recommendation.
- 4Transaction prompt. A route from implied suitability straight to paperwork.
It also flagged the performance figure as unsourced and the missing past performance warning, and explained why the disclaimer in the middle does not cure any of it.
Client communications
Before anything goes out to a client.
The Advice Boundary Checker
SAFE AdaptReads any client-facing message and marks the point where information becomes a personal recommendation. Explains each crossing, then rewrites the message without them.
For advisers, paraplanners and the people who review their files
FCA Financial Promotions Compliance Checker
SAFE Fine-TuneChecks promotions and client communications against the fair, clear and not misleading standard. Flags unbalanced risk, misuse of regulatory status and scarcity pressure.
For compliance officers approving marketing and social posts
SAFE Liability Shield
SAFE Fine-TuneScans a message for accidental advice, implied guarantees and commitments the sender did not intend to make.
For anyone writing to clients under professional obligations
Claims and evidence
For anything that asserts a fact or an outcome.
The SAFE Evidence & Claims Validator™
V.E.R.I.F.Y.™Scores a claim on how well it is evidenced, then shows what would need to be true, or produced, to support it. Works across finance, healthcare and legal.
For marketing sign-off and claims review
SAFE Reality-Check & Overconfidence Detector™
V.E.R.I.F.Y.™Finds the certainty a message is claiming but has not earned: guarantees, absolutes, and outcomes presented as settled when they are not.
For reviewing AI-drafted content before it is relied on
Governance and risk
The documents and registers a regulator asks to see.
The SAFE AI Governance Architect™
SAFE™ ExecuteDrafts the governance layer: AI use policies, risk registers, staff rules and the oversight arrangements behind them. Produces documents your own governance function then owns and approves.
For heads of risk and governance setting up AI controls
Smart S.A.F.E. Risk Assessment
SAFE™ AdaptScores an AI use case and shows which part of it carries the risk, rather than returning a single number with no working.
For triaging AI requests coming in from the business
Smart S.A.F.E. Compliance Checklist Generator
SAFE™ ExecuteTurns an AI use case into a checklist of the controls and evidence it needs, so the gaps are visible before anything is signed off.
For building a repeatable review process
Data and prompting
What goes into the model in the first place.
SAFE Data Sanitiser & Redaction Firewall™
SAFE™ SecureStrips identifiers out of text before it reaches an AI tool, and shows what was removed and why. Catches the combinations that still identify someone after the name has gone.
For teams pasting real work into general AI tools
SAFE Prompt Generator
SAFE™ SecureRewrites a risky prompt into one that does the same job without exposing client data. Shows what to strip, why it matters, and where a placeholder is protecting nothing.
For anyone using AI on real client matters
What these are not
These tools support your judgement. They do not replace it, and using one does not make a communication compliant. Nothing here is legal or compliance advice. Rule positions change, so check any reference against the current FCA Handbook, and put anything client-facing through your firm’s own sign-off.