AI Assistant Memory: The Chinese Wall Bankers Need
AI assistant memory is the feature that makes an assistant feel like it knows you: your peers, your format, what you cared about last week. In a bank, memory without walls is a problem, because some things one person knows must never reach another.
Banks solved this for people decades ago with information barriers, the Chinese walls between deal teams, research and trading. AI assistants need the same idea built in, and finance leaders should ask for it before anything else.
Memory is what makes an assistant an analyst
An assistant without memory answers every question as if it were the first. You re-explain your peer set, your preferred format and the decision the board made last quarter, every time. That is a search engine with better grammar.
With memory, the second week feels different. You say "keep it to five bullets" once, and it sticks. It knows which three banks you benchmark against. The Monday brief is shaped by what you asked about last Monday. That is the moment an assistant starts saving real time.
**Intelligence answers the question. Memory answers it the way you need it.**
Fences, not freedom
The risk is the same thing that makes memory useful: it carries context forward. If the assistant that drafted a director's notes on a confidential restructuring also answers her colleague in another team, the wall that compliance built can leak through a chat window.
The answer is not less memory. It is fenced memory. Full freedom inside clear limits: what one person shares stays with that person, what one deal team discusses stays with that team, and nothing is used to answer someone else.
Regulators expect this kind of discipline. MAS's AI risk management guidelines, issued on 7 October 2026, keep institutions accountable for the third-party AI they use. OJK's 2025 AI governance guide for banks asks for safe operation and effective risk management.
**An AI's memory should respect the same walls your people do.**
Memory is now portable, and that cuts both ways
Most assistants now let you see, edit and copy what they remember about you. That makes switching easy. A new assistant can know your role, peers and preferences on day one, without months of re-training.
It also means no assistant can rely on lock-in. If what it remembers can be pasted into a rival in a minute, the only reason to stay is that it does the work better. For buyers, that is good news. For vendors, it is the right pressure.
**When memory travels, quality is the only moat left.**
Check the defaults, not the brochure
Before you let an assistant remember anything, read the defaults. Assistant Benchmark's 28 September 2026 comparison of privacy policies found that two of the three assistants it reviewed allow training on eligible interactions by default.
A default is a decision someone else made for you. For a bank, the safe defaults are simple: no training on client conversations, memory kept per person, and a way to view, correct or delete what is stored. Each person's Rapi™ is their own, and conversations are never used to answer anyone else.
**Read the default setting. That is the policy you actually agreed to.**
Four questions about memory
Ask any vendor four things. Whose conversations can my memory reach? Can I see and delete what it remembers? Is anything I share used to train a model or answer someone else? And if I leave, can I take my memory with me?
Good answers are short and specific. For how to test the rest of an assistant, see how to test an AI analyst. The goal is an assistant that remembers everything you want it to, and nothing it shouldn't.
Sources
- Monetary Authority of Singapore: MAS Sets Out Supervisory Expectations on Responsible AI Adoption by Financial Institutions (7 October 2026)
- Otoritas Jasa Keuangan: Peluncuran Tata Kelola Kecerdasan Artifisial Perbankan Indonesia (SP 67/GKPB/OJK/IV/2025)
- Assistant Benchmark: Reports and analysis
Common questions
- What is AI assistant memory?
- AI assistant memory is what an assistant keeps between conversations: your preferences, your role, the people and peers you mention, and past instructions. It lets the assistant shape answers to how you work without you repeating yourself, and most assistants now let you view, edit or delete what they remember.
- What is a Chinese wall for AI in a bank?
- It is the AI version of a bank's information barrier: what one person or deal team shares with an assistant stays with them and is never used to answer someone on the other side. In practice that means memory kept per person or team, no training on client conversations, and clear controls to view and delete what is stored.
- Can I move my AI assistant's memory to another tool?
- Often, yes. Many assistants let you view and copy what they remember, which a new assistant can use to learn your role and preferences quickly. That makes switching easier and puts pressure on every vendor to keep users through better work rather than lock-in.
