AI Agents · Lesson

Risk and Compliance Guardrails

Position sizing limits, prohibited instrument checks, and audit logging.

Risk and Compliance Guardrails is a free AI Agents lesson on CoddyKit — lesson 3 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the AI Agents learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.

Why Financial Agents Need Hard Guardrails

A financial agent that can recommend any trade without restrictions could expose users to catastrophic losses or regulatory violations. Unlike a human advisor, an agent will execute its recommendation logic consistently — which means a bug or misconfiguration affects every user.

Hard guardrails prevent the worst outcomes.

Position Size Limits

Limit how much of the portfolio the agent can recommend putting into any single asset. A common rule: no single position should exceed 20% of total portfolio value. This prevents catastrophic concentration risk.

MAX_POSITION_PCT = 0.20   # 20% of portfolio max per asset

def check_position_size(ticker: str, trade_usd: float,
                        portfolio_value: float) -> dict:
    proposed_pct = trade_usd / portfolio_value
    if proposed_pct > MAX_POSITION_PCT:
        return {
            'allowed': False,
            'reason': (
                f'Position size {proposed_pct:.1%} exceeds limit {MAX_POSITION_PCT:.1%}. '
                f'Max trade for this portfolio: USD {portfolio_value * MAX_POSITION_PCT:.0f}'
            )
        }
    return {'allowed': True, 'position_pct': round(proposed_pct, 4)}

if __name__ == '__main__':
    print('Large trade:', check_position_size('TSLA', 30000, 100000))
    print('Small trade:', check_position_size('TSLA', 10000, 100000))

Prohibited Instruments List

Maintain a list of instruments the agent is not allowed to recommend: penny stocks, highly leveraged ETFs, sanctioned securities, or instruments not suitable for retail investors.

PROHIBITED_TICKERS = {
    'UVXY',   # VIX leveraged ETF — too volatile
    'TVIX',   # Delisted leveraged VIX
    'JDST',   # 3x leveraged gold miners short
    'NUGT',   # 3x leveraged gold miners long
    'YANG',   # 3x China short
}

PROHIBITED_TYPES = {'penny_stock', 'leveraged_3x', 'inverse_volatility'}

def check_prohibited_instrument(ticker: str) -> dict:
    if ticker.upper() in PROHIBITED_TICKERS:
        return {
            'allowed': False,
            'reason': f'{ticker} is on the prohibited instruments list'
        }
    return {'allowed': True}

if __name__ == '__main__':
    print("Checking UVXY:", check_prohibited_instrument('UVXY'))
    print("Checking AAPL:", check_prohibited_instrument('AAPL'))

Pattern-Day-Trader Rule Check

In the US, an account with less than USD 25,000 is flagged as a Pattern Day Trader (PDT) if it makes 4+ round-trip trades in 5 business days. The agent must check trade frequency before recommending same-day trades.

from datetime import datetime, timedelta, timezone

def check_pdt_rule(account_value_usd: float,
                   recent_trades: list[dict]) -> dict:
    if account_value_usd >= 25_000:
        return {'at_risk': False, 'reason': 'Account above PDT threshold'}

    cutoff = datetime.now(timezone.utc) - timedelta(days=5)
    recent = [
        t for t in recent_trades
        if datetime.fromisoformat(t['timestamp']) > cutoff
        and t.get('is_day_trade', False)
    ]

    if len(recent) >= 3:
        return {
            'at_risk': True,
            'day_trades_in_5_days': len(recent),
            'reason': (
                f'Account has {len(recent)} day trades in the last 5 days. '
                f'A 4th would trigger the PDT restriction.'
            )
        }
    return {'at_risk': False, 'day_trades_in_5_days': len(recent)}

if __name__ == '__main__':
    now = datetime.now(timezone.utc)
    trades = [
        {'timestamp': (now - timedelta(days=1)).isoformat(), 'is_day_trade': True},
        {'timestamp': (now - timedelta(days=2)).isoformat(), 'is_day_trade': True},
        {'timestamp': (now - timedelta(days=3)).isoformat(), 'is_day_trade': True},
    ]
    result = check_pdt_rule(account_value_usd=10000, recent_trades=trades)
    print('PDT rule check:', result)

Volatility-Based Trade Size Scaling

More volatile assets carry more risk. Scale position sizes inversely with volatility: if an asset has twice the volatility of the baseline, suggest half the position size.

import yfinance as yf, numpy as np

BASELINE_VOL = 0.15   # 15% annualized vol — S&P 500 average

def volatility_scaled_size(ticker: str, base_allocation_usd: float) -> dict:
    hist = yf.Ticker(ticker).history(period='1y')['Close'].pct_change().dropna()
    vol = float(hist.std() * np.sqrt(252))

    scaling_factor = BASELINE_VOL / max(vol, 0.01)
    adjusted_usd   = base_allocation_usd * scaling_factor

    return {
        'ticker':             ticker,
        'annualized_vol':     round(vol, 4),
        'scaling_factor':     round(scaling_factor, 3),
        'adjusted_size_usd':  round(adjusted_usd, 2),
        'note': 'High-vol asset: position scaled down' if scaling_factor < 1 else 'Normal sizing'
    }

Audit Logging Every Recommendation

Every trade recommendation — whether accepted or rejected — must be logged with full context: timestamp, user ID, recommended action, guardrail checks run, and outcome. This creates a complete audit trail for compliance review.

import json, hashlib, time, logging

audit_logger = logging.getLogger('financial_agent.audit')
audit_logger.setLevel(logging.INFO)

def audit_log(user_id: str, action: str, ticker: str,
              amount_usd: float, guardrail_results: list[dict],
              approved: bool, reason: str):
    log_entry = {
        'timestamp':         time.time(),
        'user_id':           user_id,
        'action':            action,
        'ticker':            ticker,
        'amount_usd':        amount_usd,
        'guardrails':        guardrail_results,
        'approved':          approved,
        'reason':            reason,
        'entry_id':          hashlib.md5(f'{user_id}{time.time()}'.encode()).hexdigest()[:12]
    }
    audit_logger.info(json.dumps(log_entry))

if __name__ == '__main__':
    import sys
    audit_logger.addHandler(logging.StreamHandler(sys.stdout))
    audit_log('user_42', 'BUY', 'AAPL', 5000.0,
               [{'check': 'position_size', 'allowed': True}],
               True, 'Within limits')

Disclaimer Injection

Any financial output from the agent must include a disclaimer that it is not personalized financial advice. Inject this automatically — never rely on the LLM to add it consistently.

FINANCIAL_DISCLAIMER = (
    '\n\n---\n'
    'DISCLAIMER: This output is generated by an AI agent and is for '
    'informational purposes only. It does not constitute personalized financial advice. '
    'Past performance does not guarantee future results. '
    'Please consult a licensed financial advisor before making investment decisions.'
)

def inject_disclaimer(response: str) -> str:
    if 'DISCLAIMER' not in response:
        return response + FINANCIAL_DISCLAIMER
    return response

if __name__ == '__main__':
    response = 'AAPL is currently trading at $150 with strong upward momentum.'
    print(inject_disclaimer(response))

Running All Guardrails in Sequence

Create a single guardrail_check() function that runs all checks in order and returns a pass/fail decision with the reason. The agent calls this before finalizing any recommendation.

def guardrail_check(user_id: str, ticker: str, action: str,
                    amount_usd: float, portfolio_value: float,
                    account_value: float, recent_trades: list[dict]) -> dict:
    checks = []

    # 1. Prohibited instruments
    prohibited = check_prohibited_instrument(ticker)
    checks.append({'check': 'prohibited_instrument', **prohibited})
    if not prohibited['allowed']:
        audit_log(user_id, action, ticker, amount_usd, checks, False, prohibited['reason'])
        return {'approved': False, 'reason': prohibited['reason'], 'checks': checks}

    # 2. Position size
    size_check = check_position_size(ticker, amount_usd, portfolio_value)
    checks.append({'check': 'position_size', **size_check})
    if not size_check['allowed']:
        audit_log(user_id, action, ticker, amount_usd, checks, False, size_check['reason'])
        return {'approved': False, 'reason': size_check['reason'], 'checks': checks}

    # 3. PDT rule
    if action == 'day_trade':
        pdt = check_pdt_rule(account_value, recent_trades)
        checks.append({'check': 'pdt_rule', **pdt})
        if pdt['at_risk']:
            return {'approved': False, 'reason': pdt['reason'], 'checks': checks}

    audit_log(user_id, action, ticker, amount_usd, checks, True, 'All guardrails passed')
    return {'approved': True, 'checks': checks}

Suitability Check by Risk Profile

Match the recommendation to the user's stated risk tolerance. A conservative investor should not be recommended high-beta growth stocks; an aggressive investor can handle higher volatility.

RISK_PROFILES = {
    'conservative':   {'max_beta': 0.8,  'max_vol_pct': 15},
    'moderate':       {'max_beta': 1.2,  'max_vol_pct': 25},
    'aggressive':     {'max_beta': 2.0,  'max_vol_pct': 50},
}

def suitability_check(risk_profile: str, ticker_beta: float,
                       ticker_vol_pct: float) -> dict:
    limits = RISK_PROFILES.get(risk_profile, RISK_PROFILES['moderate'])
    if ticker_beta > limits['max_beta']:
        return {
            'suitable': False,
            'reason': f'Beta {ticker_beta:.2f} exceeds {risk_profile} limit {limits["max_beta"]}'
        }
    if ticker_vol_pct > limits['max_vol_pct']:
        return {
            'suitable': False,
            'reason': f'Volatility {ticker_vol_pct:.1f}% exceeds {risk_profile} limit {limits["max_vol_pct"]}%'
        }
    return {'suitable': True}

if __name__ == '__main__':
    print('Conservative check:', suitability_check('conservative', ticker_beta=1.5, ticker_vol_pct=20))
    print('Aggressive check:  ', suitability_check('aggressive', ticker_beta=1.5, ticker_vol_pct=20))

Rate-Limiting Recommendations

An agent that generates trade recommendations too frequently encourages overtrading, which increases costs and risk. Apply a cooldown period between recommendations for the same asset.

import time

last_recommendation: dict[str, float] = {}  # ticker -> timestamp
COOLDOWN_HOURS = 24

def check_recommendation_cooldown(ticker: str) -> dict:
    last = last_recommendation.get(ticker, 0)
    elapsed_hours = (time.time() - last) / 3600

    if elapsed_hours < COOLDOWN_HOURS:
        remaining = COOLDOWN_HOURS - elapsed_hours
        return {
            'allowed': False,
            'reason': f'Recommendation cooldown: {remaining:.1f} hours remaining for {ticker}'
        }
    return {'allowed': True}

def record_recommendation(ticker: str):
    last_recommendation[ticker] = time.time()

if __name__ == '__main__':
    print('Before recommending:', check_recommendation_cooldown('AAPL'))
    record_recommendation('AAPL')
    print('Right after recommending:', check_recommendation_cooldown('AAPL'))

Stop-Loss and Drawdown Circuit Breakers

Beyond individual position limits, implement a portfolio-level circuit breaker: if the total portfolio loses more than X% in a single day, the agent stops making new buy recommendations until a human reviews the situation.

MAX_DAILY_DRAWDOWN_PCT = 0.05   # 5% daily loss limit

def check_daily_drawdown(portfolio_start_of_day: float,
                          portfolio_current: float) -> dict:
    daily_loss = (portfolio_current - portfolio_start_of_day) / portfolio_start_of_day
    if daily_loss < -MAX_DAILY_DRAWDOWN_PCT:
        return {
            'circuit_breaker': True,
            'daily_loss_pct':  round(daily_loss * 100, 2),
            'reason': (
                f'Portfolio down {abs(daily_loss)*100:.1f}% today. '
                f'Agent suspended — human review required.'
            )
        }
    return {'circuit_breaker': False, 'daily_loss_pct': round(daily_loss * 100, 2)}

if __name__ == '__main__':
    print('Small loss day:', check_daily_drawdown(100000, 98000))
    print('Big loss day:  ', check_daily_drawdown(100000, 92000))

What does the Pattern Day Trader (PDT) rule restrict?

The PDT rule is a US regulatory requirement that financial agents operating in the US market must respect. Understanding it prevents agents from inadvertently triggering margin account restrictions.

Risk and Compliance Guardrails Recap

Financial agent guardrails include: position size limits, prohibited instruments list, PDT rule checking, volatility-scaled sizing, suitability matching to risk profile, audit logging every recommendation, and disclaimer injection in all outputs.

Always run all guardrails before finalizing any trade recommendation.

Frequently asked questions

Is the “Risk and Compliance Guardrails” lesson free?

Yes — the full text of “Risk and Compliance Guardrails” is free to read here on the web, and the AI Agents course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the AI Agents course, upgrade to CoddyKit PRO.

What will I learn in “Risk and Compliance Guardrails”?

Position sizing limits, prohibited instrument checks, and audit logging. You practise AI Agents with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.

Do I need any experience to start AI Agents?

No prior experience is required. AI Agents on CoddyKit is structured for beginners through advanced learners; this is — lesson 3 of 4, so you can start here or from the beginning and move at your own pace.

How long does the “Risk and Compliance Guardrails” lesson take?

Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.

Can I write and run code in this AI Agents lesson?

Yes. Every AI Agents lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.

All lessons in this course

← Back to AI Agents