FuzzingLabs has launched FuzzForge, an AI agent-powered platform for automated security research.1 Where traditional fuzzing generates random inputs and monitors for crashes, FuzzForge uses autonomous agents that reason about code structure, track coverage, and adapt strategies during execution.2
Method
Classical fuzzing is effective but blind. It has no understanding of what the code does or which inputs are likely to expose vulnerabilities. FuzzForge puts AI agents in the loop. Each agent reads source code, analyzes execution traces, formulates hypotheses about where bugs may exist, and generates targeted inputs to test them.
Application
Smart contracts manage substantial assets with no reversal mechanism. The attack surface is broad: EVM bytecode, Cairo contracts, Solana programs, Move modules, each with distinct edge cases.
FuzzingLabs has spent years building fuzzing tools for these environments. Their Cairo fuzzer, Starknet analysis tools, and blockchain-specific harnesses represent accumulated knowledge about where these systems fail. FuzzForge makes this autonomous—running continuously across every commit, surfacing findings for human review.