21 risk controls every trading bot needs

Not strategies. Controls: the code that decides whether a bug, a lost connection, a poisoned headline or an exhausted model quota costs you a little or costs you the account. Six groups; each control comes from a documented failure.

Size and exposure

Maximum order size — Is there a hard cap per order, and in which unit is it validated: quote currency, base asset, or percent of balance?
Execution paths that skip the cap — Do all order-sending paths go through the same risk check, or are there shortcuts such as rebalance, emergency close, batch or a newer strategy that bypass it?
Total exposure and leverage — Besides the per-order cap, is there a limit on total exposure: the sum of open positions, the number of positions, or the maximum leverage?

Brakes and stops

Global circuit breaker — Is there a kill switch that stops the whole bot after a daily loss, a drawdown, or a number of consecutive failures?
Orphaned stop-loss — If the bot restarts or loses its connection, do open positions keep a stop order at the exchange, or does the stop only exist in the bot's memory?
Orderly shutdown — When told to stop (signal, Ctrl+C, command), does it cancel open orders and record the open positions, or does it just exit?

Connection and operation

Websocket and data reconnection — Does it reconnect on its own when the websocket or API drops, with bounded retries and without duplicating subscriptions? Does it back off on rate limits (HTTP 429)?
Duplicate-process lock — Does it prevent two instances from running at the same time, with a lockfile, a PID file, or a semaphore?
Heartbeat and alerts — Does it notify the user (Telegram, email, webhook) when it fails, stops or loses its data feed, and does it expose a heartbeat or health status that can be watched from outside?

Accounting and reconciliation

Position reconciliation with the exchange — At startup and periodically, does it compare its own positions with the exchange's and resolve the differences?
Fees and funding in the accounting — Does it subtract maker and taker fees and, on perpetuals, funding payments when computing profit and position size?
Order rejection handling — What happens when the exchange rejects an order or fills it partially: bounded retry, log entry, alert, or silent loop?
Auditable log — Does it keep a persistent record of every decision and order: time, reason, size, and the exchange's response?

Security

API key permissions — Do the docs ask for an API key without withdrawal rights and with IP restriction, does the code work with such a key, and does the code ever call withdrawal or transfer endpoints?
Secrets handling — Do API keys come from environment variables or a secret store, never from the code or versioned files, and do they stay out of the logs?
Authenticated remote control — If it is controlled through Telegram, Discord or a web UI, does it obey only authorised users (id allow-list, token, password), or anyone who finds the bot?

AI and learning

The model cannot override the brakes — If an AI model makes the trading decision, does that decision still pass through the deterministic controls (cap, breaker, exposure), or can the model set the size and skip the limits?
Model output validation — Does it validate the model's answer (format, existing symbol, side, quantity, price within range) before turning it into an order, and reject malformed or invented answers?
Learns from its mistakes — Does it record the outcome of each decision and use it to correct itself (parameter tuning, trade memory, prompt or strategy changes)? If it self-corrects, does the change go through a review, a backtest or a limit before trading with it?
Model poisoning — If it feeds news, tweets or third-party messages into the model, are they treated as data (isolated, limited, with no power to order), or can an external text dictate a trade?
Behavior without model quota or response — If the model does not answer, runs out of quota, or returns garbage, does the bot stop, hold its positions, or keep trading blind? Is there a daily cap on token cost?