CVE Intelligence
High Insecure deserialization

CVE-2025-9906 — Arbitrary Code execution in Keras Safe Mode

The Keras Model.load_model method can be exploited to achieve arbitrary code execution, even with safe_mode=True. One can create a specially crafted .keras model archive that, when loaded via Model.load_model, will trigger arbitrary code to be executed. This is achieved by crafting a special config.json (a file within the .keras archive) that will invoke keras.config.enable_unsafe_deserialization() to disable safe mode. Once safe mode is disable, one can use the Lambda layer feature of keras, which allows arbitrary Python code in the form of pickled code. Both can appear in the same archive. Simply the keras.config.enable_unsafe_deserialization() needs to appear first in the archive and the Lambda with arbitrary code needs to be second.

Published Sep 19, 2025 · Updated Oct 19, 2026 · Sources: NIST National Vulnerability Database, CVE Program (cve.org) e CISA ADP, GitHub Security Advisories · CNA: Google

Who exploits it and how urgent it is

A 0–100 score combining confirmed exploitation, ransomware and APT use, public exploits, EPSS and severity.

No exploitation signal in the sources consulted so far.
How the score was calculated
  • EPSS 0.2% (30-day exploitation probability)+1
  • Technical severity (high)+6

The score orders the patch queue; it does not replace your asset context (internet exposure, criticality, compensating controls).

Triage

Is it exploited, how likely is exploitation, what does it touch, and how severe do the sources call it.

Exploitation
Not confirmed
Not in CISA KEV — an absence of record, not proof it is unexploited.
EPSS FIRST
0.2%
ahead of 10% of scored CVEs · Oct 20, 2026
Affects
Keras
Keras and 1 more
CVSS base
7.3
High · 4 source(s)

CISA SSVC assessment CISA

Three decision points CISA publishes for the CVEs it assesses (SSVC 2.0.3). A stakeholder decision, not a severity score.

Exploitation
None
none · proof-of-concept · active
Automatable
No
can an attacker script the kill-chain steps
Technical impact
Total
partial · total control of the vulnerable component

Remediation

Fixes and mitigations documented by the sources — linked to the original advisory.

  • Pip Keras: fixed in version 3.11.0 — github.com (GitHub Security Advisories)

First 24 hours

Ordered from the record’s own fields — exposure first, because you cannot patch what you have not found. A rule-generated starting point.

  1. Identify exposed assets running the affected vendor/product/version combinations (Keras Keras, Keras-team Keras-team/keras, Pip Keras).
  2. Apply the fixed version or the patch documented under Remediation.
  3. Exploitation requires user interaction: tighten email/web filtering and warn affected users.
  4. Search available logs for exploit probes matching the vulnerability class (insecure deserialization), errors and authentication anomalies.

Affected scope

Vendor, product and versions as the sources word them.

VendorProductVersionsStatusSource
Keras Keras ≥ 3.0.0 and < 3.11.0 Vulnerable NIST National Vulnerability Database
Keras-team Keras-team/keras ≥ 3.0.0 and < 3.11.0 Vulnerable CVE Program (cve.org) e CISA ADP
Pip Keras < 3.11.0 Vulnerable GitHub Security Advisories

Attack characteristics

The CVSS vector, decoded. It describes the attack, not your exposure to it.

Attack vector
Local
Attack complexity
Low
Privileges required
Low
User interaction
Required
Scope
Unchanged
Confidentiality
High
Integrity
High
Availability
High

Every base score collected

Sources score independently and may disagree; each row says who scored it and under which version.

ScoreVersionSeverityExpl.ImpactScored by
8.6 CVSS 4.0 High — — cve-coordination@google.com via NIST National Vulnerability Database
8.6 CVSS 4.0 High — — Google via CVE Program (cve.org) e CISA ADP
7.3 CVSS 3.1 High 1.3 5.9 nvd@nist.gov via NIST National Vulnerability Database
7.3 CVSS 3.1 High — — GitHub, Inc. via GitHub Security Advisories
CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:U/C:H/I:H/A:H (score shown at the top: nvd@nist.gov)

Weakness MITRE

Weakness class according to MITRE’s CWE catalog.

CWE-502 · Deserialization of Untrusted Data

Other CVEs with this weakness →

EPSS over time FIRST

Oct 19, 2026 → Oct 20, 2026 · 2 reading(s) · max 0.20%

Detection

Read off the CVSS vector and the weakness class. Starting points, not tested rules.

  • Local/physical vector: focus on endpoint telemetry (EDR) from affected hosts.
  • Monitor serialized objects in requests and unexpected processes spawned by the runtime.
endpoint telemetry (EDR)application logs

Timeline

What happened to this CVE, newest first.

  • 2026
  • Fix documented
    Oct 19, 2026 · GitHub Security Advisories
  • Affected products added · pip/keras
    Oct 19, 2026 · GitHub Security Advisories
  • Affected products added · keras-team/keras
    Oct 19, 2026 · CVE Program (cve.org) e CISA ADP
  • SSVC assessment updated
    Oct 19, 2026 · CVE Program (cve.org) e CISA ADP
  • Record published
    Oct 19, 2026
  • Affected products added · keras/keras
    Oct 19, 2026 · NIST National Vulnerability Database
  • CVSS changed · 7.3
    Oct 19, 2026 · NIST National Vulnerability Database
  • Source status changed · → Analyzed
    Oct 19, 2026 · NIST National Vulnerability Database
  • Record created in this database
    Oct 19, 2026 · Alertas do Telegram
  • 2025
  • CVE published
    Sep 19, 2025 · Google

Source activity

Repeated readings are counted rather than listed.

  • 2× References added · Oct 19, 2026 – Oct 19, 2026
  • 1× Official description updated · Oct 19, 2026 – Oct 19, 2026

References

7 on the record

Elsewhere on this site

Keras
every CVE for this vendor
Insecure deserialization
same class
CWE-502
other pages naming this weakness

Related CVEs (same products)

  • CVE-2026-12570 Denial of Service via HDF5 Shape Bomb in keras.models.load_model() in keras-team/keras Aug 10, 2026
  • CVE-2026-9335 Improper Handling of HDF5 ExternalLinks in keras-team/keras Aug 2, 2026
  • CVE-2026-12484 Unsafe Deserialization in keras.layers.TorchModuleWrapper.from_config Jul 19, 2026
  • CVE-2026-12482 Path Traversal via Symlink Name Validation Bypass in keras-team/keras Jul 14, 2026
  • CVE-2026-12481 Deserialization of Untrusted Data in keras-team/keras Jul 3, 2026
  • CVE-2026-12480 Arbitrary HDF5 File Read via Virtual Dataset Bypass in keras-team/keras Jul 1, 2026

Not in any source we poll

Listed rather than left blank: an empty field and an unmeasured one look identical on screen, and only one is a reason to look elsewhere.

  • No confirmed IOCs, IP addresses, domains, file hashes, or malware artifacts supplied.
  • No organization-specific asset inventory, compensating-control status, or patch deployment evidence supplied.
  • No exploit packet captures, log samples, or incident case IDs supplied.

Answered from this record

Text assembled only from the fields above — no additional claims.

What should defenders know first?

CVE-2025-9906 is "Arbitrary Code execution in Keras Safe Mode", rated High. It affects Keras (Keras); versions are in the scope table. It is not in CISA KEV, which does not rule out exploitation. CVSS 7.3 describes technical severity. EPSS 0.20% estimates the probability of exploitation in 30 days; neither replaces your asset context. First verify whether exposed or business-critical assets run the affected versions, apply vendor fixes or mitigations, and preserve logs for review.

Comments

Moderated technical discussion. Read the community guidelines.

No comments yet.

Sign in or create an account to comment and suggest corrections.