What is trading?

Trading is risky. Your capital is at risk.

CROSS-CUTTING: COURSE 1 | LESSON 6

Building a safe AI workflow: privacy, verification habits, and what never to delegate

Learning objectives

Apply the privacy rules for AI tools: what may never enter a prompt, and how to sanitise trading data before pasting.
Install the verification habits from this course as a single pre-decision checklist.
Draw the delegation line: which parts of the trading process AI may support, and which must never leave your hands.

From techniques to a system

The first five lessons gave you techniques: honest expectations (X1.1), the prompt pattern (X1.2), the journal analyst (X1.3), digestion workflows (X1.4) and strategy sparring (X1.5). Techniques used ad hoc drift toward carelessness — one rushed morning you'll paste something you shouldn't, or skip verification because the answer looked right. This closing lesson turns the techniques into a system: three short rulebooks you set up once and stop renegotiating.

Rulebook 1 — privacy: what never enters a prompt

Assume anything you type into an AI tool may be stored, reviewed for quality purposes, or — depending on your settings and provider — used to train future models. You don't need to be paranoid about a pasted trade list; you need a hard line, and here it is.

Never paste, under any circumstances:

Account numbers, login credentials, passwords, or API keys — for your trading account, bank, email, anything.
Identity documents or their contents: passport numbers, national ID, proof-of-address details.
Full names paired with financial details — yours or anyone else's.
Card numbers, wallet addresses paired with your identity, or account balances tied to your name.

Safe to paste (and all the skills in this course need nothing more):

Instruments, directions, entry/exit prices, dates and times.
Trade results in R multiples (a habit from X1.3 that keeps balances out of prompts entirely).
Your written trading rules, economic-calendar rows, and public documents like central bank statements.

The test is simple: could this text harm you if it appeared on a public website? If yes, it doesn't enter a prompt — no exceptions for "just this once", and no trusting a chat interface's delete button as an undo for a pasted password (change the password instead). Also be deliberate about where you use these tools: prefer official apps and websites of major providers, check your provider's data/training settings once, and be very wary of unknown "AI trading" apps that ask you to connect your brokerage account. A legitimate use of an LLM never requires access to your money.

Rulebook 2 — verification: one checklist to rule them all

Scattered through this course were verification habits. Consolidated, they fit on an index card. Before any AI output influences a trading decision:

Numbers — every rate, price, percentage and date checked against the primary source (central bank site, official release, your platform, your own journal export).
Quotes — verbatim-searchable in the document you pasted. Unfindable quote → discard the claim, distrust the neighbours.
Recency — if you didn't paste it in this session, assume it predates the model's cutoff and treat it as stale.
Negations — reread the original sentence behind any summary line that would change your bias (the dropped-"not" problem from X1.4).
Grounding — every conclusion cites the rows, rules or sentences it rests on. "Which claims are computed from my data, and which are interpretation?" remains the best closing question.
Mirroring — did the answer just agree with the bias in your question? Re-ask neutrally once and compare.

Six checks, two to ten minutes. If that feels heavy, invert it: an unverified AI "fact" that survives into a live trade can cost you a month of profits, and you won't even know which belief was the fake one.

Example — the checklist catching a real failure mode: Your morning briefing says: "German CPI at 12:00 is the day's key EUR event (forecast 2.4% vs previous 2.8%)." Check 1: your pasted calendar row says forecast 2.4%, previous 2.6% — the AI mutated a number while reformatting. Small slip, but your "big disinflation surprise" narrative was built on it. Thirty seconds of checking deleted a false belief before it priced itself into your trading day.

Rulebook 3 — the delegation line

The deepest safety rule isn't about data or facts — it's about decisions. Draw the line once, in writing, and keep it:

AI may support (with verification): summarising and diffing research you provide · analysing your journal · stress-testing written rules · explaining concepts · drafting checklists and plan documents · asking you hard questions.

Never delegated, ever: the decision to enter or exit a trade · position size · where your stop goes · whether to trade at all today · risk-limit overrides ("the AI made a good case for risking more") · and any belief about what price will do next.

The pattern behind the banned list: these are the decisions where accountability is the point. An AI is never accountable — it carries none of your risk, remembers nothing tomorrow, and delivers wrong answers in the same warm tone as right ones. The moment "the assistant thought it looked good" enters your decision chain, you've built a responsibility-laundering machine: something to blame that can't be blamed. Regulators put it more formally — AI output is not investment advice; it is unvetted, machine-generated text. Nothing an assistant tells you moves any responsibility off your shoulders, and nothing in this course is investment advice either.

This is also why every skill file in our library hard-codes the same rails into its prompt: no buy/sell recommendations, flag missing data instead of inventing it, show which user numbers each conclusion rests on, remind the user this is not financial advice. The rails aren't decoration — they're Rulebook 3, embedded where you can't forget it.

Your complete safe workflow

Putting the whole course together, a healthy AI-assisted trading week looks like:

Daily: paste the calendar → Morning Macro Briefing skill → verify numbers → note the day's event risk in your plan.
Per trade: your own rules decide entry, size, stop — AI untouched. Optionally, the Pre-Trade Risk Check skill audits your arithmetic and plan-compliance before you click (checking your numbers, never blessing the trade).
Per closed trade: two minutes with the Post-Trade Review Coach skill while it's fresh.
Monthly: journal export → Trade-Journal Analyst skill → answer its five questions in writing → if rules change, one round with the Strategy Stress-Tester.

Everything flows one direction: your data and decisions in, structure and questions out. The moment the arrow reverses — the AI supplying facts you didn't verify or decisions you didn't make — you're outside the system. That's the whole discipline. It fits on a card, and it's worth more than any prompt in this course.

Key takeaways

Privacy line: credentials, account numbers, identity documents and name-linked finances never enter a prompt; R multiples and sanitised trade data are all any workflow here needs.
Run the six-point verification checklist — numbers, quotes, recency, negations, grounding, mirroring — before AI output touches any decision.
The delegation line is absolute: entries, exits, sizing, stops and "should I trade today" are never the AI's, because accountability can't be delegated.
AI output is not investment advice — it's unvetted machine-generated text, and every skill in our library embeds that framing in its rules.
A safe workflow flows one way: your verified data and your decisions in; structure, summaries and hard questions out.