Skip to main content

Policy Controls

A library of 100+ potential policy controls for AI agent wallets

Comprehensive reference for spending controls in SpendSafe policy layer. Custom policies, contribution ideas, organised by category.

About This Document​

Purpose: Catalog potential policy controls for AI agent wallet management

Status: Living document—contributions welcome!

Current Implementation: See "Currently Implemented ✅" section below

Contributing: Found a missing control or have a better approach? Open an issue or submit a PR

Why Policy Controls Matter​

AI agents with direct wallet access can:

  • Drain funds through runaway spending
  • Fall victim to prompt injection attacks
  • Violate compliance requirements
  • Create liability for operators

Policy controls solve this by enforcing rules before transactions are signed—without custodying keys.

How to Use This Document​

For Developers​

  • Find controls that match your use case
  • Check implementation status
  • Review the API documentation for implementation guidance
  • Start with Phase 2 controls for near-term wins

For Contributors​

  • Pick an unimplemented control
  • Open an issue describing your approach
  • Submit a PR with implementation + tests
  • Update this document with "✅ Implemented"

For Product Teams​

  • Identify controls needed for your use case
  • Prioritise based on risk/value
  • Share feedback on missing controls

Currently Implemented ✅​

Basic Spending Limits​

  • Per-Transaction Limit: Maximum amount per single transaction
  • Daily Limit: Maximum total spending per 24-hour period (midnight reset)
  • Hourly Limit: Maximum total spending per hour (UTC hour boundaries)

Recipient Controls​

  • Recipient Whitelist: Only allow transactions to pre-approved addresses

Velocity Controls​

  • Transaction Frequency: Maximum X transactions per time period (e.g., 3 per hour)

Spending & Volume Controls​

Time-Based Limits​

  • Hourly Limit: Maximum spend per hour
  • Weekly Limit: Maximum spend per 7-day rolling window
  • Monthly Limit: Maximum spend per 30-day rolling window
  • Quarterly/Annual Limits: Longer-term spending caps

Velocity Controls​

  • Transaction Frequency: Max X transactions per time period (e.g., 10 per hour)
  • Cooldown Period: Minimum time between transactions
  • Burst Protection: Allow short bursts but limit sustained high-volume activity
  • Rate Limiting by Recipient: Max X transactions to same address per period

Amount-Based Rules​

  • Minimum Transaction Amount: Prevent dust/spam transactions
  • Tiered Limits: Different limits based on transaction size brackets
  • Progressive Limits: Limits that decrease as more is spent
  • Accelerating Limits: Require additional checks above certain thresholds

Recipient Controls​

Whitelists & Blacklists​

  • Recipient Whitelist: Only allow transactions to pre-approved addresses
  • Recipient Blacklist: Block transactions to specific addresses
  • Contract Interaction Whitelist: Only allow interaction with approved smart contracts
  • DNS/ENS Whitelist: Only allow transactions to verified domain names

Recipient Analysis​

  • New Address Delays: Require waiting period for first transaction to new address
  • High-Risk Address Blocking: Block known scam/hack addresses (Chainalysis integration)
  • Contract Verification: Require contracts to be verified on Etherscan
  • Recipient Reputation Scoring: Allow/deny based on on-chain reputation

Asset & Chain Controls​

Multi-Asset Policies​

  • Asset-Specific Limits: Different limits for ETH vs USDC vs other tokens
  • Stablecoin vs Volatile: Different rules for stable vs volatile assets
  • NFT Transfer Policies: Special rules for NFT transactions
  • Token Approval Limits: Max approval amounts for ERC-20 approvals

Multi-Chain Policies​

  • Chain-Specific Limits: Different limits per blockchain
  • Bridge Transaction Rules: Special policies for cross-chain bridges
  • Preferred Chains: Encourage/require using specific chains
  • Chain Risk Scoring: Higher scrutiny for newer/riskier chains

Temporal Controls​

Time Windows​

  • Operating Hours: Only allow transactions during business hours
  • Weekend Restrictions: Different limits on weekends
  • Holiday Freezes: Disable transactions during holidays
  • Timezone-Based Rules: Adjust rules based on user's timezone

Scheduling​

  • Scheduled Transactions: Pre-approve transactions for future execution
  • Recurring Payments: Automated recurring transfers with limits
  • Embargo Periods: Block transactions during sensitive periods

Security Controls​

Multi-Party Approval​

  • Two-Person Rule: Require approval from second party above threshold
  • Multi-Signature: Require M-of-N approvals for large amounts
  • Supervisor Approval: Route large transactions to human supervisor
  • Delayed Execution: Enforce waiting period allowing cancellation

Fraud Detection​

  • Anomaly Detection: Flag unusual patterns (ML-based)
  • Geo-Fencing: Block if transaction originates from unexpected location
  • Device Fingerprinting: Block if different device suddenly used
  • Behavioral Analysis: Compare to agent's normal transaction patterns

Risk Management​

  • Progressive Trust: Start with low limits, increase as agent proves reliable
  • Circuit Breakers: Auto-pause if unusual activity detected
  • Emergency Freeze: Admin ability to instantly freeze agent
  • Rollback/Revert: Ability to reverse recent transactions (where possible)

Compliance & Regulatory​

KYC/AML​

  • Transaction Reporting: Auto-report transactions above threshold
  • Suspicious Activity Flagging: Flag potentially suspicious patterns
  • Sanctions Screening: Check against OFAC and other sanctions lists
  • Source of Funds: Track and verify origin of funds

Audit & Reporting​

  • Detailed Logging: Record all transaction attempts (success and failure)
  • Compliance Reports: Generate reports for regulators
  • Tax Reporting: Prepare transaction summaries for tax purposes
  • Real-Time Alerts: Notify stakeholders of significant events

Jurisdictional​

  • Regional Restrictions: Different rules based on legal jurisdiction
  • Currency Controls: Respect capital controls in certain countries
  • Prohibited Activities: Block transactions related to prohibited goods/services

Agent-Specific Controls​

Identity & Authentication​

  • Agent Identity Verification: Verify agent is who it claims to be
  • API Key Rotation: Require periodic credential rotation
  • Session Timeouts: Require re-authentication after inactivity
  • MFA Requirements: Multi-factor auth for sensitive operations

Purpose & Context​

  • Transaction Purpose Tags: Require agents to tag transaction purpose
  • Business Logic Validation: Ensure transaction makes business sense
  • Invoice Matching: Match transactions to approved invoices
  • Budget Allocation: Enforce budget categories (marketing, ops, etc.)

Learning & Adaptation​

  • Confidence Scores: Require higher confidence for larger amounts
  • Trial Periods: Limited capabilities when agent is new
  • Performance-Based Limits: Adjust limits based on agent success rate
  • Feedback Loops: Learn from human approvals/rejections

Technical Controls​

Gas & Fees​

  • Maximum Gas Price: Prevent overpaying for gas
  • Gas Limit Caps: Prevent excessive computation
  • Priority Fee Controls: Limit MEV/priority fees
  • Fee Optimization: Require use of gas optimization services

Smart Contract Interaction​

  • Function Whitelist: Only allow calling specific contract functions
  • Parameter Validation: Validate smart contract call parameters
  • Reentrancy Protection: Prevent reentrancy attacks
  • Upgrade Restrictions: Block interaction with upgradeable contracts

Network & Infrastructure​

  • RPC Endpoint Restrictions: Only use approved RPC providers
  • Mempool Privacy: Require private mempool for sensitive transactions
  • MEV Protection: Use Flashbots or similar MEV protection
  • Simulation Requirements: Require transaction simulation before execution

Advanced Policy Features​

Conditional Logic​

  • If-Then Rules: "If amount > X, then require Y"
  • Combined Conditions: Multiple criteria must be met
  • Exception Handling: Override rules in specific circumstances
  • Policy Inheritance: Agents inherit policies from groups/organizations

Dynamic Policies​

  • Market-Responsive: Adjust limits based on market volatility
  • Liquidity-Aware: Consider available liquidity before approving
  • Time-Decay: Limits that change throughout the day
  • Event-Triggered: Policies that activate on specific events

Organizational​

  • Department Budgets: Enforce departmental spending limits
  • Project-Based: Different limits for different projects
  • Cost Center Allocation: Track spending by cost center
  • Hierarchical Policies: Company-wide → Department → Agent

Data & Privacy​

Privacy Controls​

  • Data Minimization: Only collect necessary transaction data
  • Encryption: Encrypt sensitive policy and transaction data
  • Right to Deletion: Allow data deletion after retention period
  • Anonymization: Anonymize data for analytics

Transparency​

  • Policy Disclosure: Agents understand what rules apply to them
  • Explanation System: Explain why transactions were blocked
  • Appeal Process: Allow challenging of policy decisions
  • Audit Trail: Complete history of policy changes

Integration & Extensibility​

External Systems​

  • Oracle Integration: Use price oracles for USD limits
  • Banking Integration: Coordinate with traditional bank accounts
  • Accounting System Sync: Push transactions to QuickBooks/Xero
  • Notification Systems: Alert via Slack/email/SMS

Custom Rules Engine​

  • Rule Builder UI: No-code interface for creating policies
  • Policy Templates: Pre-built policies for common use cases
  • Policy Versioning: Track changes to policies over time
  • A/B Testing: Test different policies simultaneously

AI/ML Enhancements​

  • Predictive Blocking: Block likely fraudulent transactions proactively
  • Smart Recommendations: Suggest policy improvements
  • Pattern Recognition: Identify emerging fraud patterns
  • Natural Language Policies: "Block transactions over $1000 on weekends"

Emergency & Recovery​

Incident Response​

  • Panic Button: Instantly freeze all agents
  • Gradual Recovery: Slowly restore access after incident
  • Post-Incident Review: Mandatory review after security events
  • Lessons Learned: Update policies based on incidents

Disaster Recovery​

  • Backup Policies: Fallback rules if primary system fails
  • Manual Override: Human ability to override in emergencies
  • Degraded Mode: Reduced functionality during system issues
  • Recovery Procedures: Clear process for restoring normal operations

Implementation Priority​

Phase 1 (MVP - Current)​

  • ✅ Per-Transaction Limit
  • ✅ Daily Limit
  • ✅ Basic audit logging

Phase 2 (Near-Term)​

  • Recipient whitelist/blacklist
  • Hourly/weekly limits
  • Transaction frequency controls
  • Asset-specific limits

Phase 3 (Medium-Term)​

  • Multi-party approval
  • Anomaly detection
  • Compliance reporting
  • Time-window restrictions

Phase 4 (Long-Term)​

  • ML-based fraud detection
  • Dynamic policy adjustment
  • Advanced organizational features
  • Full regulatory compliance suite

Document Version: 1.0 Last Updated: 2025-10-20 Maintainer: Wollit Team