Where Does AI Proposal End and Human Authority Begin in a Trade?

An authority map assigns every trade decision to either AI proposal or human approval before an order reaches the market. Signal generation, position sizing, order type, modification, and exit can begin as AI recommendations, but execution starts only after the trader authorizes the complete order.

In April 1970, Apollo 13 was losing oxygen after an onboard tank explosion, and the crew’s planned Moon landing had become a survival problem. Astronauts Jim Lovell, Jack Swigert, and Fred Haise had limited power, rising carbon dioxide, and no room for casual improvisation.

Mission Control in Houston calculated procedures. Flight Director Gene Kranz coordinated specialists responsible for separate systems. The crew performed the approved steps in space, where every switch movement had consequences.

Lovell and Jeffrey Kluger document this division of responsibility in Lost Moon. Expertise was distributed, but authority remained explicit. People knew who could diagnose, who could propose, who could approve, and who had to perform each action.

That structure offers a useful model for AI-assisted trading. A system may detect an opportunity and calculate an order. The trader still needs a clear boundary between analysis and irreversible action.

Six decisions inside one trade

A trade signal can look like a single decision: buy or sell. In practice, it contains at least six.

1. Signal generation identifies a possible setup from market data and defined rules.
2. Position sizing determines how much capital the idea could expose.
3. Order type controls how the trade may enter, such as a market or limit order.
4. Execution sends the approved instruction to the broker or exchange.
5. Modification changes a live order, stop, target, or position size.
6. Exit closes some or all of the position.

Each decision changes risk differently. A reasonable signal paired with an oversized position remains an unreasonable trade. A valid entry can become unacceptable after a stop is widened. A cautious limit order can turn into immediate exposure if changed to a market order.

This is why an approval screen needs more than a direction and ticker. Before authorizing execution, the trader should see the proposed quantity, order type, entry conditions, stop logic, exit conditions, and total exposure alongside existing positions.

Where AI may propose

AI can help with the work that happens before commitment. It can scan defined markets, identify setups, compare a proposed trade with stated risk limits, and queue the complete instruction for review.

It can also explain its reasoning in testable terms. For example, the proposal might identify the condition that triggered the signal, the price assumption used for sizing, and the event that would invalidate the setup. Those details give the trader something concrete to inspect.

Position sizing deserves its own check. Suppose a stop is close to the entry price. A quantity calculated from that narrow distance may become much larger than expected. The signal can remain technically valid while the account-level exposure becomes unacceptable. That is the distinction explored in AI Trading Risk Management: Why Daniel Rejected a Valid but Oversized Order.

The proposal stage should end with a complete, readable order ticket. It should not quietly continue into execution.

Where human authorization begins

Human authorization begins before the first order is sent. Approval should cover the whole proposed package, including size and order type, rather than the broad idea alone.

The same boundary applies after entry. A modification creates a new risk decision. Increasing quantity, moving a stop farther away, changing the order type, adding another position, or re-entering after an exit should return to the approval queue.

An exit also needs an explicit policy. A trader may pre-authorize a defined protective stop as part of the original order. A later discretionary exit proposed by AI requires fresh approval. The system should distinguish those cases visibly, because “close according to the approved stop” and “close now for a newly detected reason” have different authority histories.

Timing does not erase the boundary. If an alert arrives while the trader cannot review it, the proposal should remain unexecuted or expire according to a rule the trader chose beforehand. What Happens When a Trade Alert Arrives Before You Can Review the Risk? examines that specific failure point.

Build the map before the next signal

Write six rows in a trading journal: signal, size, order type, execution, modification, and exit. Beside each row, record whether AI may propose, whether the trader must approve, and what happens when approval does not arrive.

Then test the difficult cases. Can the system add to a position without another decision? Can it widen a stop? Can it replace an expired order? Can it re-enter after a loss? Can a previously approved instruction execute hours later under different market conditions?

Apollo 13 returned safely because Mission Control and the crew worked through defined responsibilities under pressure. Your authority map serves the same narrower mechanism: analysis can be distributed, while consequential actions remain assigned to a named decision-maker.

For an approval-gated trading assistant, that decision-maker is the trader. The next practical step is simple: open the current order workflow and mark the exact click, API call, or rule where a proposal becomes a live instruction.