OpenVM CVE-2026-62333

Exploitation requires the attacker to control proof generation (act as a malicious prover) in an OpenVM-based system, and applies only to proofs that use persistent memory, since the memory Merkle root binds VM memory state across continuation segment boundaries. The concrete trigger is a Merkle expansion row set with parent_height = 0 (or below) while expand_direction != 0, which pre-fix code failed to forbid, allowing a fabricated below-leaf subtree whose digest swaps change the committed root with no memory-bus operation. Affected code paths are OpenVM before 1.7.0 in the 1.x line and the affected 2.0 and 2.1 prereleases; deployments not using persistent memory, or already on 1.7.0 / 2.0.0-rc.3 / 2.1.0-preview or later, are not exploitable. Updating to a fixed version is the remediation stated by the vendor.

This is a zero-knowledge proof soundness flaw, and the available signals point to a genuine but scoped priority rather than a mass-exploitation emergency. The CVSS 4.0 vector (AV:N/AC:L/AT:N/PR:N/UI:N) scores 8.7 and is driven entirely by integrity: VI:H with VC:N and VA:N, consistent with a bug that forges the committed memory root without leaking data or affecting availability. The vuln.today decision label is 'this_week' (not act_now/emergency). No EPSS probability and no CISA KEV status were provided, so there is no evidence of active or widespread exploitation; the absence of EPSS/KEV should not be read as a clean bill of health, only as missing signals. A public counterexample demonstrating the attack ships in the fix commit's test suite (build_below_leaf_swap_subtree), so exploitability is well-understood even without a weaponized exploit. The real caveat is reachability: this is a malicious-prover soundness break, so the 'attacker' must control proof generation, and only OpenVM deployments using persistent memory across continuation segment boundaries are affected. For those users this undermines the core trust guarantee of their proofs and is high priority; for the broader population it is low-footprint. The 'Information Disclosure' tag conflicts with the integrity-only impact and appears mislabeled.

An attacker acting as the prover in an OpenVM system that uses persistent memory crafts a malicious execution trace placing a Merkle expansion row at parent_height = 0 while expand_direction != 0, fabricating a subtree below a memory leaf and swapping digests to commit a different memory root with no corresponding memory-bus operation. The forged rows still balance against the Merkle and compression buses, so the resulting proof verifies and convinces a verifier of a false memory state across continuation segments. A public PoC reference exists as a counterexample test in the fix commit; it is not verified as a standalone working exploit.