Threat Model

Surf is designed under the assumption that every layer can fail and that adversarial behaviour is the default state of open financial systems. The security model does not rely on trust in the AI, operators, or off-chain components. It is built on explicit threat modelling and deterministic containment.

The primary question we design for is not:

“Can the system make money?”

It is:

“What happens when something goes wrong, and how is loss prevented or contained?”


Core Threat Classes

1. AI Decision Risk

The AI can:

  • Misinterpret signals

  • Overfit to short-term noise

  • Propose actions that are statistically valid but operationally unsafe

  • React incorrectly under regime shifts

Mitigation:

AI has zero direct execution authority. Every action must pass the Guardian Layer, which enforces:

  • Allowlisted venues only

  • Exposure and allocation caps

  • Slippage and price-impact bounds

  • Volatility and anomaly filters

  • Cooldown and rate limits

  • Simulation-based pre-checks

If a proposal violates any invariant, it is rejected and nothing moves.


2. Smart Contract Risk

Threats include:

  • Logic bugs

  • Reentrancy

  • Oracle manipulation

  • State desynchronisation

  • Upgrade path abuse

Mitigation:

  • Vaults are isolated per user

  • Execution modules are permissioned and minimal

  • External protocol interactions are allowlisted

  • Continuous internal audits and external security reviews

  • Emergency pause and unwind paths are hard-coded

No strategy logic can arbitrarily move funds outside its authorised execution scope.


3. Liquidity and Market Structure Risk

Includes:

  • Sudden liquidity withdrawal

  • MEV and sandwich attacks

  • Oracle lag

  • Extreme volatility and gap moves

  • CLMM range breaks

  • Lending market utilisation spikes

Mitigation:

  • Pre-trade simulation and price impact checks

  • Maximum position sizing per venue

  • Dynamic slippage ceilings

  • Volatility regime detection

  • Circuit breakers that freeze execution during stress

  • Fallback to safe-state allocations

The system is built to prefer not executing over executing in degraded conditions.


4. Cross-Chain and Bridge Risk

Threats:

  • Bridge compromise

  • Message reordering

  • Replay or finality failure

  • Liquidity fragmentation

Mitigation:

  • Chain abstraction is execution-controlled, not free-routing

  • Bridges are allowlisted per asset and per route

  • Finality thresholds enforced before state transitions

  • Cross-chain moves require Guardian approval

  • Emergency asset isolation per chain

No autonomous cross-chain movement can bypass risk checks.


5. Custody and Key Management Risk

Threats:

  • Key compromise

  • Operator abuse

  • Hot wallet drain

  • Signature forgery

Mitigation:

  • User-owned vaults

  • MPC (Multi-Party Computation) signing

  • No single key can move funds

  • Deterministic signing policies

  • Hardware-backed security for operational keys

  • Role separation between planning, approval, and execution


6. Governance and Upgrade Risk

Threats:

  • Malicious upgrades

  • Governance capture

  • Backdoor insertion

Mitigation:

  • Time-locked upgrades

  • Multi-sig and MPC enforcement

  • Public audit trails

  • Emergency veto paths

  • Invariant checks that cannot be overridden by governance


Failure Containment Philosophy

Surf assumes that:

  • Markets will break

  • Oracles will lag

  • Liquidity will disappear

  • AI will be wrong

  • Smart contracts will be probed

  • Adversaries will continuously attack

The system is therefore built around containment, not optimism.

When something fails:

  1. The action is blocked by rules

  2. Capital remains in the vault

  3. State is frozen or reverted

  4. Emergency exit paths are available

  5. Human and automated monitors are triggered

  6. No cascading execution is allowed

Loss is not allowed to propagate across:

  • Users

  • Strategies

  • Chains

  • Venues


Design Invariant

The fundamental invariant of Surf is:

No component, human or AI, can move user funds outside deterministic safety boundaries.

Everything else, including performance, comes second.

This threat model is why Surf behaves differently from:

  • Strategy routers

  • Yield aggregators

  • Black-box agents

  • Custodial automation platforms

Surf is built as an execution system under adversarial assumptions, not as an optimisation script.

That is what makes it safe to let AI touch real capital.

Last updated