Risk Management in Trading, for Humans and AI Agents

Risk management in trading decides the loss before the gain. The rules are the same for a person and an AI agent. People break them under stress, and models misread prompts, so the limit has to sit where the decision-maker cannot edit it.

This guide covers the 1% rule, position sizing, stop-losses, take-profits, risk/reward and loss limits, then onchain differences, then agent controls. Building the system is covered in how to build an AI trading bot, which points here for the risk policy. A price can go to zero, and a gap can exceed a plan.

What is risk management in trading?

Risk management in trading means limiting how much you can lose on any trade, day or strategy. The core tools are position sizing, stop-loss orders, a minimum risk/reward ratio, daily loss limits and diversification. When an AI agent trades for you, the same limits apply, enforced by the platform rather than the model.

"Typical setting" means a common rule of thumb from Investopedia's guide for active traders. It is a starting habit, with no promised outcome.

Keep a rule you intend to follow, and a limit something else enforces. You can plan a 1% exit and then slide the stop. A platform cap on order size cannot be talked out of the number. Choose the loss first, derive the size from that loss, and place the exit before the entry.

The 1% rule and position sizing

The 1% rule means you risk about 1% of account value on one trade, from entry to the stop. Investopedia's risk-management FAQ puts it this way: on a $10,000 account, plan not to lose more than about $100. The same article's main text also describes capping the whole position at 1%. This page uses the risk reading, because that is what the formula calculates. The two match only when the stop is the entire position.

Investopedia's position sizing page uses the same math. As a rule of thumb it says many retail traders risk no more than about 2% of capital on one trade. Treat 1% and 2% as habits, and use the smaller one when the market is jumpy. The same page notes gap risk: price can jump past the stop, and the loss can exceed the plan.

In units, position size = (account size x risk per trade %) / (entry price minus stop price). The top is the dollars you are willing to lose. The bottom is the loss on each unit if the stop fills at the price you typed.

Illustrative example only: not a token, a price, or a stop to copy. Account $2,000. At 1%, dollars at risk are $20. The stop is 10% below the entry, so position value = $20 / 0.10 = $200. At a $1 token that is 200 units, stop at $0.90. Planned loss is $20. A gap or a thin pool can fill below $0.90. Volatile tokens need wider stops, which means smaller positions: at a 25% stop, the same $20 is an $80 position. Leave the position at $200 and widen the stop, and the trade is no longer a 1% risk.

A daily income target pushes size up and stops aside. Many active traders lose money. The CFTC and the SEC, NASAA and FINRA warn against pitches that treat a trading system as a paycheck.

Stop-losses, take-profits and risk/reward

Stop-loss orders

A stop-loss order closes a trade at a level you set before you enter. Investopedia's stop-loss page describes the listed-market version: at the stop price the order becomes a market order and fills at the next available price, so a gap can fill worse than the stop.

Onchain, a stop is a trigger order someone has to place. A buy does not carry a stop unless one is placed afterward. In a thin pool the fill can be worse than the trigger. Set the stop where the idea is wrong, then size from that distance. A stop so tight that ordinary noise hits it closes trades that were still valid. A wider stop without a smaller position is a larger bet.

Take-profit orders

A take-profit order closes a trade at the price where you will bank the gain. Set it with the stop, before you enter. On a listed market it is often a limit. Onchain it is a trigger or a limit you place yourself. Leave it unplaced, and a round trip can hand the gain back. Moving the target because the trade looks strong is the same late edit as moving the stop.

Risk/reward ratio

Risk/reward compares the stop distance with the target distance. Risk $1 to target $2 or more is the usual reading of at least 1:2. In the illustrative example, $20 at risk and a 1:2 minimum is a $40 target. On the $200 position with a 10% stop, that target sits about 20% above the entry, about $1.20 if the token is $1. The figures show the ratio. They are not a forecast.

Win rate and the ratio work together. Risking $1 to make $2 can still lose money over a series if winners are rare. Risking $1 to make $1 needs winners more often than losers. Skip ideas whose target is too close, before you enter.

Loss limits and exposure caps

Per-trade risk is one trade. Add a daily loss limit, a daily volume cap, a maximum number of open positions, and a cap on how much of the account sits in one asset or one sector.

A daily loss limit is where new trades stop for the day. Trades taken to make a loss back are where the 1% rule usually gets dropped. Pick the number before the session, often a small multiple of one full loss. A written number does nothing if you keep clicking, which is why an agent needs the same limit in software.

A daily volume cap limits how much you buy in a day, win or lose. Many small buys can each sit inside the per-trade cap and still put a large part of the account to work. The loss limit watches damage already done. The volume cap watches what you are still allowed to spend.

Three open positions is a fair planning number on a small account, and three related names can still act as one. Memecoins on one narrative, AI chip stocks, and tokenized stocks in one sector can fall together. Cap the theme as well as the ticker. Investopedia treats that diversification as a core control, beside the stop and the 1% rule.

A market-cap floor is the onchain way to stay in names that can absorb an order. Majors-only guardrails are one written setup: a floor aimed at majors and tokenized stocks, so an agent stays off the smallest contracts. A floor is a filter, not a view on whether a large token is a good trade.

Crypto risk management: what is different onchain

Crypto risk management keeps the same loss math and adds four problems a stock-and-forex guide usually skips.

Markets run 24 hours a day, seven days a week. There is no closing bell, so a daily loss limit needs a clock you choose, such as a rolling 24 hours or your own morning. An agent that trades while you sleep needs that clock even more.

A tokenized stock can trade onchain while the underlying listed stock is closed, including over a weekend. The two prices can move apart until that listed market reopens, so the stop has to allow for the gap. That is a general feature of a listed price on a 24/7 chain, not a claim about one venue.

In a thin pool your order moves the price, and the fill is worse than the quote you saw. A market-cap floor keeps you away from the thinnest markets, and a slippage limit on the order belongs with the stop, chosen before you send it. Both change which markets you touch. Neither sets the fill.

Mint authority can add supply, freeze authority can block a sell, concentrated holders can sell into you, and a honeypot blocks the sell so a stop never fires. Check those before you size. A rug check covers mint and freeze authority, holder concentration, and honeypot-style flags.

A trading wallet holds only trading funds. Savings stay in a different wallet the trading key cannot reach, so a bad trade or a leaked key is confined to the trading balance.

What changes when an AI agent trades

The rules above stay the rules. Speed, the type of mistake, and where the limit lives are what change.

An agent can send many orders a minute for as long as it runs, so the per-trade cap and the daily caps matter more than they do for a person. Fifty small buys inside a loose daily cap are an exposure problem. Set both before you turn it on.

Agents misread prompts, act on stale prices, and hallucinate tickers. A familiar symbol can be a different contract. Size, daily buy volume, and a market-cap range or allowlist have to be checked by the platform, outside the model. Agentic trading is that split: the agent researches and proposes, and account rules decide what may execute. The Claude trading bot guide puts the same rules on Claude Code.

Three modes: trade only when told, propose and wait for approval, or trade alone inside rules you set. Start with approval, and let it trade alone only after it has followed the policy on small size. Tool permissions and token allowlists are a separate checklist. The loss math on this page still applies.

Prompt injection and keys

OWASP lists prompt injection as LLM01. Untrusted text can carry instructions, and a model may follow them instead of yours. Onchain that text sits in token descriptions, websites, and posts. A page can tell an agent to ignore your policy, buy a specific mint, or ask you to paste a key. Treat market text as data. Your policy and your approval rule stay the instructions.

Use a scoped key you can revoke, and keep it out of the chat. Never share a wallet private key with a model. A trading key you can turn off is a different object from the key that moves the wallet. A prompt that asks for the private key is the attack.

Regulators' view

The SEC, NASAA and FINRA investor alert says not to rely solely on AI-generated information. It can be inaccurate, incomplete, or made up, and a chatbot can push a hurried trade. The CFTC advisory on AI trading bots says AI cannot predict the future or sudden market changes, and that fraudsters use AI language to pitch automated systems with promises of high or guaranteed returns. FINRA, on auto-trading services from unregistered entities, lists "AI can automate risk controls" among AI-washing lines. In those pitches, the line is a red flag. Automation enforces the limits you set. It does not remove the risk of the trade.

Build a kill switch

A kill switch stops new trading at once. Pause new trades, cancel open orders, and revoke the API key. You want all three, and you want to know where each one lives before you need it.

Pausing stops new ideas. It leaves a resting stop, take-profit, or scheduled buy in place, so cancel those too. A paused session can be turned back on. A revoked key cannot send another order until you create a new one on purpose. Write down the pause screen, the open-order screen, and the revoke screen, then test once: new trades stop, resting orders are gone, and the old key fails. The same three steps apply to a script you wrote yourself.

A risk policy you can give any AI agent

Paste a short policy into the agent's instructions, with numbers you have actually chosen. Edit the block if your account needs a different habit. Keep the structure: per-trade risk, a stop on every buy, a minimum ratio, a cap on open positions, a market-cap floor, a daily stop, and approval.

Risk no more than 1% of account value on any one trade, measured from entry to the stop.
Place a stop-loss on every buy.
Take a trade only when the target is at least twice the stop distance (minimum 1:2).
Hold at most 3 open positions.
Do not buy tokens below the market-cap floor I set.
After the daily loss limit is hit, stop for the day.
Ask before every trade.

The agent's instructions are guidance, and platform-enforced caps are the backstop. The model can misread the block, and prompt injection can tell it to ignore the block. When they disagree, the platform number should stand, and it should be the tighter one. Claude Trading Skills (position sizer, drawdown circuit breaker) can run the size and the daily-loss check as habits beside the policy. They do not replace a cap checked on the buy.

How Moby enforces risk for agents

Moby is one place these limits can be enforced. The caps bound the size of buys. They are checked on buys, and they do not bound how far a price can fall after you hold the token.

You set your agent's limits on max per trade, daily buy volume and market-cap range from the Moby app, and Moby checks them server-side on every buy your agent makes.

Base limits shown in the app are up to 6 trades a minute and 200 a day. Those are rate limits on top of the dollar caps. A trade the server blocks stays blocked. The agent is told not to retry it, not to split it into smaller orders, and not to switch tools to get around the cap.

Moby attaches no exit orders to a buy. A take-profit or a stop-loss exists only when the agent places it with create_order. Put that in the policy, then check the order is open. Pausing the agent blocks new trades. Reads still work, so balances and open orders stay visible while nothing new goes out. Revoke the key from the Agent tab. The agent's tools cannot withdraw, transfer, or send funds. Swaps stay in the agent wallet.

get_token_trade_brief is a one-call pre-buy read, including safety signals, plus about, flow, holders, and market. There is one agent per user, with its own wallet, separate from the main in-app wallet. Fund it with USDC. It trades Solana tokens, including tokenized stocks.

Try it with a Moby Agent

Agentic trading is the wider picture. Set up a Moby Agent when you are ready to connect.

1. Download Moby and open the Agent tab. One agent per user, with its own wallet, separate from the main in-app wallet.
2. Set max per trade, daily buy volume and market-cap range to match your risk policy.
3. Connect Codex, Claude Code (CLI) or Grok with an agent key, and choose the mode that asks you to approve each trade.
4. Fund the agent wallet with USDC. It trades Solana tokens, including tokenized stocks such as SPYx and GLDx.
5. Paste the risk policy above and ask the agent to place a stop-loss with every buy. Then confirm the stop order is open.

A 0.75% swap fee (minimum $0.90 per swap), plus onchain fees. 10% of the Moby fee back in $MOBY on every swap, 20% after your first swap, up to 30% as you climb tiers.

Agent trading on Base, BNB Chain and Robinhood Chain is coming soon.

FAQ

What is the 1% rule in trading?

Risking no more than about 1% of your account on a single trade, measured from entry to stop-loss, not the full position size. A gap can still fill worse than the stop.

How do I calculate position size?

Multiply account size by your risk percentage, then divide by the distance between entry and stop price. The result is the number of units. Investopedia's position sizing page uses those same steps.

What is a good risk/reward ratio?

Many traders look for at least 1:2, meaning the target is twice the distance of the stop. The right ratio depends on win rate: a wider target can still lose over a series if winners are rare.

Can an AI trading bot manage risk on its own?

It can follow rules, but models make mistakes. Keep caps enforced by the platform, start with trade approval, and keep a kill switch. Pasted instructions are guidance. A cap the platform checks on the buy is the backstop.

How is crypto risk management different?

Markets run 24/7, liquidity can be thin, and tokens carry contract risks, so add market-cap floors, slippage limits and safety checks. A buy does not include a stop unless you place one. Tokenized stocks can keep trading while the listed market they track is closed.

What is a trading kill switch?

A way to stop all new trading at once: pause the bot or agent, cancel open orders and revoke its keys. Learn where each control is, and test it once, before you need it.

What is the 3-5-7 rule in trading?

Definitions vary between sources. One version, from DayTrading.com, reads 3% as the most to risk on one trade, 5% as the most to risk across open positions, and 7% as a goal that winners be at least 7% larger than losers. It is one version, not a standard, and other writers define the 7 differently. Per trade it is looser than the 1% rule.