SECURITY MODEL / PUBLIC

Evidence, with
honest limits.

The rules that keep NOLEAK useful without turning security research into false certainty.

+
01 / POLICY

An observation is not an assurance

NOLEAK distinguishes evidence of an onchain signature from proof of private-key compromise. An address with no signature in a checked window is not automatically considered safe.

02 / POLICY

Every scan has a boundary

Browser scans cover a bounded number of blocks, chosen by the user or near a recent snapshot. Offchain signatures, EIP-712 messages and the history of other chains may not be observable.

03 / POLICY

A migration is never automatic

The Migration Safety module is a preparatory check on user-selected assets. It does not inventory all assets, replace transaction simulation for every protocol, authorize, sign or broadcast any transaction.

04 / POLICY

Agents are not authorities

zkAPI may provide privacy between API usage and billing, but not hide prompt contents from the model provider. IMD job acceptance does not certify the security truth of a submitted report.

05 / POLICY

Privacy is an operational choice

Public addresses reveal activity to RPC providers and researchers. For sensitive audit work, disclose only what is necessary. NEVER paste a seed phrase, private key or recovery code.

06 / POLICY

Cryptography is still evolving

Potential AI-accelerated attacks are research concerns, not confirmed breaks of ECDSA or ML-DSA. NOLEAK tracks established facts and labels speculative threat models.

Protocol references

EIP-7702 ↗zkAPI ↗IMD documentation ↗Safe documentation ↗