Skip to content

Security: FrostedFlaming0/BasedAI

Security

SECURITY.md

Security

Reporting vulnerabilities

Do not open public issues for security problems.

Email: security@basedai.network (PGP key: see docs/pgp.txt)

Or use the Immunefi program: https://immunefi.com/bounty/basedai/ (live after mainnet launch).

Bug bounty scope

Critical (up to $250,000):

  • Loss of user funds
  • Theft of staked BASED
  • Incorrect emission distribution that drains the supply

High (up to $50,000):

  • Brain ownership theft
  • Validator equivocation that bypasses slashing
  • Receipt double-spend

Medium (up to $10,000):

  • Griefing miners or validators out of legitimate emissions
  • Permission escalation in governance

Out of scope:

  • Issues on testnet
  • Frontend issues that don't affect contract security
  • Social engineering of token holders
  • DoS via gas exhaustion (except where it leads to fund loss)

Audits

This repository is pre-audit. The audit candidate must be identified by an explicit commit or tag and accompanied by:

  • docs/audit/audit-scope.md
  • docs/audit/threat-model.md
  • docs/audit/privilege-matrix.md
  • docs/audit/deployment-manifest.template.json or a network-specific manifest generated from it

Highest-priority audit surfaces are the custom contracts (BrainNFT, StakingVault, ScoringRegistry, ComputeUnitMarket, RewardDistributor, SubnetRegistry, BasedGovernor) and deployment scripts. BrainNFTL2 is small and mostly OpenZeppelin, but its immutable bridge/remote-token wiring is security-critical.

Known limitations

  • Validator scoring is heuristic. Before mainnet, use a non-public eval set, add collusion/Sybil monitoring, and treat score roots as routing/slashing inputs rather than as proof of model correctness.
  • Equivocation fraud proofs require off-chain monitoring. A future version should consider permissionless proof rewards funded from slashed stake.
  • The HTTP gateway/aggregator path is the testnet transport. The libp2p module remains a stub and is explicitly out of mainnet readiness unless separately completed and audited.
  • Public testnet must still validate live Ink bridge message passing end to end: L1 deposit -> L2 mint -> L2 withdrawal -> L1 release.

Operational security

The bootstrap guardian may hold emergency pause capability for ComputeUnitMarket and timelock cancellation capability only. It cannot:

  • Mint BASED outside the emission schedule
  • Modify the emission schedule
  • Seize Brain NFTs
  • Override scoring results
  • Withdraw user balances or staked funds

Protocol administration is transferred to the timelocked governor during deployment; the deployer renounces bootstrap admin roles before the script completes.

There aren't any published security advisories