CVE-2026-92702 Details

A vulnerability exists in Cocos AI versions through 0.8.2, specifically within the intra-handshake attested TLS (aTLS) implementation for AMD SEV-SNP. The issue arises because the verification process does not properly enforce the freshness of attestation when the expected reportData is nil, empty, or omitted. This flaw allows the verifier to accept unrelated or stale evidence that is not bound to the current connection, potentially leading to unauthorized trust in an attestation context. Although non-empty reportData is validated, the absence of this data can be exploited by a relying party that uses the verification as a trust or authorization decision.

Users are advised to upgrade to Cocos AI version 0.9.0 or later, where this vulnerability has been addressed by migrating from intra-handshake attestation to post-handshake attestation. For those who cannot upgrade, a workaround is to explicitly require a current-connection reportData value and fail closed before accepting verification if the value is absent, empty, or malformed.