CVE-2025-69418: OpenSSL: OCB tail bytes left unencrypted in low-level API

MEDIUM
Published January 27, 2026
CISO Take

CVE-2025-69418 is a flaw in OpenSSL's low-level OCB encryption routines, where on hardware-accelerated builds the last 1-15 bytes of a message that is not a multiple of 16 bytes can be left in cleartext and excluded from the authentication tag. Real-world exposure is very narrow: the standard EVP interface is not affected, TLS does not use OCB, FIPS modules are unaffected, and the issue is only reachable by applications calling CRYPTO_ocb128_encrypt() or CRYPTO_ocb128_decrypt() directly with unaligned lengths in a single call. The data does not suggest urgency: it is not in CISA KEV, there is no public exploit or Nuclei template, EPSS is not available, and OpenSSL rates it Low (CVSS 4.0, local vector, high complexity). Treat it as routine patch hygiene: apply the January 2026 OpenSSL fixes for 3.6, 3.5, 3.4, 3.3, 3.0 and 1.1.1 where available, and prioritize any custom code that encrypts model artifacts, embeddings or inference payloads with the raw OCB API.

Sources:NVDopenssl-library.orgATLAS

What is the risk?

Low overall risk. Exploitation requires local access and high attack complexity, and the vulnerable code path is only used by applications that bypass EVP and call the low-level OCB functions directly. No known exploitation, no public exploit, and not in CISA KEV. Impact is limited to partial confidentiality and integrity loss of the final 1-15 bytes of a message. The CVE is not AI-specific; OpenSSL is a shared cryptographic dependency under many ML stacks. Confidence: HIGH on the vendor assessment, LOW on the number of affected AI deployments, since dependent counts are unknown.

How does the attack unfold?

Initial Access
Attacker obtains local or storage-level access to data encrypted with the low-level OCB API on a hardware-accelerated build.
Exploitation
Attacker reads the unencrypted trailing 1-15 bytes and modifies them, since they are not covered by the authentication tag.
Impact
Small portions of model artifacts or inference data are leaked or tampered with without detection.
AML.T0037

What systems are affected?

PackageEcosystemVulnerable RangePatched
OpenSSL——No patch

Do you use OpenSSL? You're affected.

How severe is it?

CVSS 3.1
4.0 / 10
EPSS
0.1%
chance of exploitation in 30 days
Higher than 3% of all CVEs
Exploitation Status
No known exploitation
Sophistication
Advanced

What is the attack surface?

AVACPRUISCIA
AVLocal
ACHigh
PRNone
UINone
SUnchanged
CLow
ILow
ANone

What should I do?

1 step
  1. 1) Inventory OpenSSL versions across ML base images, containers and build pipelines (3.6, 3.5, 3.4, 3.3, 3.0 and 1.1.1 are vulnerable; 1.0.2 and the FIPS modules are not). 2) Apply the patched releases from the OpenSSL advisory of 27 January 2026 (openssl-library.org/news/secadv/20260127.txt) or vendor backports. 3) Search internal code for direct calls to CRYPTO_ocb128_encrypt and CRYPTO_ocb128_decrypt; migrate to EVP-based OCB or an AEAD such as AES-GCM or ChaCha20-Poly1305. 4) If direct usage exists, re-encrypt any data produced with non-block-aligned lengths on hardware-accelerated builds and validate integrity. 5) Rebuild and redeploy container images so the updated library is picked up.

What does CISA's SSVC say?

DecisionTrack
Exploitationnone
AutomatableNo
Technical Impactpartial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Data LeakageFramework

Which compliance frameworks are affected?

This CVE is relevant to:

ISO 42001
A.7.4-Quality of data for AI systems
NIST AI RMF
MANAGE 2.3-Procedures to respond to and recover from unknown risks

Frequently Asked Questions

What is CVE-2025-69418?

CVE-2025-69418 is a flaw in OpenSSL's low-level OCB encryption routines, where on hardware-accelerated builds the last 1-15 bytes of a message that is not a multiple of 16 bytes can be left in cleartext and excluded from the authentication tag. Real-world exposure is very narrow: the standard EVP interface is not affected, TLS does not use OCB, FIPS modules are unaffected, and the issue is only reachable by applications calling CRYPTO_ocb128_encrypt() or CRYPTO_ocb128_decrypt() directly with unaligned lengths in a single call. The data does not suggest urgency: it is not in CISA KEV, there is no public exploit or Nuclei template, EPSS is not available, and OpenSSL rates it Low (CVSS 4.0, local vector, high complexity). Treat it as routine patch hygiene: apply the January 2026 OpenSSL fixes for 3.6, 3.5, 3.4, 3.3, 3.0 and 1.1.1 where available, and prioritize any custom code that encrypts model artifacts, embeddings or inference payloads with the raw OCB API.

Is CVE-2025-69418 actively exploited?

No confirmed active exploitation of CVE-2025-69418 has been reported, but organizations should still patch proactively.

How to fix CVE-2025-69418?

1) Inventory OpenSSL versions across ML base images, containers and build pipelines (3.6, 3.5, 3.4, 3.3, 3.0 and 1.1.1 are vulnerable; 1.0.2 and the FIPS modules are not). 2) Apply the patched releases from the OpenSSL advisory of 27 January 2026 (openssl-library.org/news/secadv/20260127.txt) or vendor backports. 3) Search internal code for direct calls to CRYPTO_ocb128_encrypt and CRYPTO_ocb128_decrypt; migrate to EVP-based OCB or an AEAD such as AES-GCM or ChaCha20-Poly1305. 4) If direct usage exists, re-encrypt any data produced with non-block-aligned lengths on hardware-accelerated builds and validate integrity. 5) Rebuild and redeploy container images so the updated library is picked up.

What systems are affected by CVE-2025-69418?

This vulnerability affects the following AI/ML architecture patterns: model serving, training pipelines, custom encrypted artifact storage, inference services.

What is the CVSS score for CVE-2025-69418?

CVE-2025-69418 has a CVSS v3.1 base score of 4.0 (MEDIUM). The EPSS exploitation probability is 0.13%.

What is the AI security impact?

Affected AI Architectures

model servingtraining pipelinescustom encrypted artifact storageinference services

Compliance Controls Affected

ISO 42001: A.7.4
NIST AI RMF: MANAGE 2.3

What are the technical details?

Original Advisory

Issue summary: When using the low-level OCB API directly with AES-NI or<br>other hardware-accelerated code paths, inputs whose length is not a multiple<br>of 16 bytes can leave the final partial block unencrypted and unauthenticated.<br><br>Impact summary: The trailing 1-15 bytes of a message may be exposed in<br>cleartext on encryption and are not covered by the authentication tag,<br>allowing an attacker to read or tamper with those bytes without detection.<br><br>The low-level OCB encrypt and decrypt routines in the hardware-accelerated<br>stream path process full 16-byte blocks but do not advance the input/output<br>pointers. The subsequent tail-handling code then operates on the original<br>base pointers, effectively reprocessing the beginning of the buffer while<br>leaving the actual trailing bytes unprocessed. The authentication checksum<br>also excludes the true tail bytes.<br><br>However, typical OpenSSL consumers using EVP are not affected because the<br>higher-level EVP and provider OCB implementations split inputs so that full<br>blocks and trailing partial blocks are processed in separate calls, avoiding<br>the problematic code path. Additionally, TLS does not use OCB ciphersuites.<br>The vulnerability only affects applications that call the low-level<br>CRYPTO_ocb128_encrypt() or CRYPTO_ocb128_decrypt() functions directly with<br>non-block-aligned lengths in a single call on hardware-accelerated builds.<br>For these reasons the issue was assessed as Low severity.<br><br>The FIPS modules in 3.6, 3.5, 3.4, 3.3, 3.2, 3.1 and 3.0 are not affected<br>by this issue, as OCB mode is not a FIPS-approved algorithm.<br><br>OpenSSL 3.6, 3.5, 3.4, 3.3, 3.0 and 1.1.1 are vulnerable to this issue.<br><br>OpenSSL 1.0.2 is not affected by this issue.

Exploitation Scenario

An organization built a custom service that encrypts model checkpoints and inference logs using the raw OpenSSL OCB API on AES-NI hardware, passing buffers whose length is not a multiple of 16 in a single call. An attacker with local or storage-level access to the encrypted artifacts reads the trailing bytes in plaintext and flips them without invalidating the authentication tag, for example altering a trailing metadata field or a final record. The impact is limited to those few bytes, so the attack mainly enables subtle tampering or small information leakage rather than full compromise.

Weaknesses (CWE)

CWE-325 — Missing Cryptographic Step: The product does not implement a required step in a cryptographic algorithm, resulting in weaker encryption than advertised by the algorithm.

Source: MITRE CWE corpus.

CVSS Vector

CVSS:3.1/AV:L/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N

Timeline

Published
January 27, 2026
Last Modified
October 15, 2026
First Seen
October 15, 2026

Related Vulnerabilities