AI at Blockaid
Scaling exploit detection with AI
Blockaid's agent harness lets frontier models reason about onchain environments like a real security engineer: pulling bytecode, tracing funds, and pattern-matching across contexts.
Architecture
AI that reasons like a security researcher
A real-time ML model runs on every transaction across 30+ chains, shipping verdicts in milliseconds. Cases above a certain risk threshold are investigated by agents running in our research harness.
01 Real-time model
Trained on data no blockchain indexer can provide
Trained on data no blockchain indexer can provide
Our model weighs hundreds of features spanning calldata, simulation results, approval scope, and behavioral history, returning a scored verdict in milliseconds.
02 Investigative agents
Identical calldata, opposite intent
Identical calldata, opposite intent
A drainer and a legitimate withdrawal can look the same onchain. Telling them apart requires context no single transaction carries, so we have our agents go and get it.
Models running in our harness can resolve proxies and admin roles, pull counterparty history, and work backwards from the real-time classifier feature-by-feature. When the signal was wrong, that correction goes back into the real-time model.
No buzzwords, just real world results
97%
Researcher agreement-rate180k
Transactions reviewed / month
Detect, understand, and protect against fraud, scams, exploits, and financial risks with Blockaid
