GnuPG CVE-2025-68973

Exploitation requires a vulnerable GnuPG to process an attacker-crafted OpenPGP ASCII-armored input through the de-armor path (armor_filter in g10/armor.c); the flaw is in the buffer-copy loop that fills the decode output. Concretely limiting factors: AV:L means the malicious armored data must be processed locally (e.g., a file, message, or piped input handed to gpg), AC:H means the specific buffer state that surfaces the double-increment/uninitialized-read must be reliably reached (non-trivial), and in practice a victim must actually run gpg against the attacker-supplied armored data. No non-default feature toggle is required - the armor decoder is a standard code path - but the high complexity and local processing requirement meaningfully constrain real-world exploitation.

The CVSS base score is 7.0 (HIGH), but the vector tells a more measured story: AV:L (local) and AC:H (high attack complexity) mean this is not a remotely-triggerable, point-and-click flaw - it requires GnuPG to locally process attacker-supplied data and the memory-handling condition to be reliably hit. The exploitation-probability signals strongly reinforce low urgency: EPSS is 0.02% (4th percentile), it is NOT listed in CISA KEV, and no public exploit is flagged in the input. The counterweight is that this flaw received significant public security-research attention (the gpg.fail/memcpy writeup, the 39C3 CCC talk 'To Sign or Not to Sign', oss-security, and HN threads), so technical detail is widely available even if no weaponized PoC is tracked. Note a conflicting metadata signal: the CVSS vector claims full-triad impact (C:H/I:H/A:H) while the tag set labels this 'Information Disclosure' - the source-diff root cause (a double-increment in the ASCII-armor copy loop leaking uninitialized buffer memory) is most consistent with a confidentiality/memory-disclosure primary impact, so the I:H/A:H ratings may be generous. Also note a PR discrepancy: NVD assigns PR:L, but supplying a crafted armored message to a victim's gpg realistically needs no attacker privileges (PR:N) - verify intended threat model with the vendor. Net: real but low-priority; patch on the normal maintenance cycle rather than as an emergency.

An attacker crafts a malicious OpenPGP ASCII-armored message and delivers it to a victim, who decrypts or verifies it with a vulnerable gpg. The double-increment bug in the armor decoder (g10/armor.c) advances the output index twice per iteration, causing uninitialized heap buffer contents to be emitted into the decoded output - a memory-disclosure condition. No public exploit is flagged in the input, though the flaw was detailed in public research (gpg.fail/memcpy, 39C3); AV:L/AC:H means exploitation is local and non-trivial to trigger reliably.