Native Solana Audits

Native programs are about as bare as Solana gets: a process_instruction entrypoint, a hand-written dispatch table, and every check the program author wrote themselves. A lot of the most important code on the chain is built this way for the efficiency it buys, tighter compute, smaller binaries, fewer dependencies. The trade-off is that nothing protects you from yourself, which is why we tend to find more bugs in native programs than in framework code.

There is no single way to write native code, so we adapt to however you built it. The serialization you chose, the account-validation approach you settled on, the dispatch style you prefer, we read the program on its own terms rather than against a template. We are as flexible as you need us to be.

In practice that means reading every instruction handler closely, starting from the routing layer where an attacker-controlled discriminator byte first lands, then through ownership checks, signer verification, and raw deserialization. The bugs usually come from a single missing line, a check present in one handler but quietly absent in a near-identical one, or a length that was never validated before it indexed a slice.

Native is essentially the same coding and auditing experience as a Pinocchio program, just with even less between you and the runtime. We treat the program’s structure as untrusted and look for the smallest set of accounts an attacker could supply that still satisfies every explicit check while violating the implicit ones.