Skip to content
Noroxi
CVE-2025-9906· NVD / CVE Program· CNA Google

Arbitrary Code execution in Keras Safe Mode

Noroxi assessment

Priority: Monitor (action score 34/100)

  • The sources we track (CISA KEV, Exploit-DB, Metasploit, Nuclei, GitHub) show no public exploit code or exploitation in the wild.
  • Probability of exploitation in the next 30 days: 0.20% (FIRST EPSS).
  • Someone with local access to the system can trigger it. A low-privileged account is enough. No special conditions are required; it is repeatable.
  • Fixed version: Keras-team Keras 3.11.0; azl3 keras 3.3.3-3 on Azure Linux 3.0 3.3.3-4.

Low priority for now. Upgrade to Keras-team Keras 3.11.0; azl3 keras 3.3.3-3 on Azure Linux 3.0 3.3.3-4 or later.

Generated automatically from the record's data: CVSS (NVD and CNA), CISA KEV and SSVC, FIRST EPSS, exploit sources, fixed versions. How we calculate it

Primary sources: NVD · CVE.org · FIRST EPSS

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.

HighCVSS 8.6 · v4.0—No exploit Fix available
Published
Sep 19, 2025
Updated
Jun 17, 2026
EPSS
0.2% · 9th percentile

Report tools

JSON

Action score

34

Monitor

Low priority for now.

CVSS
34 / 40 · 8.6 / 10
CISA KEV
0 / 30 · Not listed
EPSS
0 / 30 · 0.2%

CISA SSVC decision

Exploitation
none
Automatable
no
Technical impact
total

Vulnrichment: CISA's decision-tree inputs.

CNA vs NVD score

NVD
—
CNA · Google
8.6
NVD has not scored this yet; the score shown is the CNA’s.

CVSS:4.0/AV:L/AC:L/AT:N/PR:L/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/AU:Y/R:A

The score given by the assigning authority versus NVD’s independent score. A gap means the severity is contested.

Get told when this record changes

  • Not in KEV: no record of exploitation in the wild
  • No public exploit code on record
  • A fix has been released

If any of these changes, we'll tell you in the daily email digest.

Follow CVE-2025-9906

When you'll hear from us

  • when it enters CISA KEV
  • when exploit code appears or is weaponized
  • when the record is updated (including a fixed version)
  • when EPSS rises 20 points in a week

One daily email digest (06:00 UTC); connect a webhook for KEV listings and weaponized exploits instantly.

Your free account also gets you

Running WordPress? Scan your site free with WP Lens

  • Stack alerts: add the products and versions you run, get the new CVEs that affect them
  • Uptime monitoring: 1 site, instant email when it goes down
  • Add your domain: subdomains and takeover risk scanned every week (1 domain)
  • 10 scans an hour on the security tools (3 a day without an account), an API key and MCP
  • All your scans in one panel: findings, what changed in the last scan, a report per domain or per client (PDF, share link)
All free tools

Sign up with email · Have an account? Sign in

Free account, no card. Unfollow any time.

Don't want an account? Get "Exploited this week" by email every MondayDon't want an account? RSS feed for keras records

Running WordPress? Scan your site free with WP Lens

Is your version affected? Check it

Noroxi analysis

What does it mean in your environment?

The assessment above is generated automatically from the record's data; it doesn't know your environment. Our team verifies on your systems whether this vulnerability is actually exploitable and which mitigation is enough.

We use this product, ask for help

Affected systems

Is your version affected?

Pick the product and type the version you run. It's compared with this record's version ranges in your browser; the version isn't sent anywhere.

VendorProduct
keraskeras

Affected versions

NVD version ranges (for catalog products). Add the product to your stack with a version and matching uses these.

  • keras keras3.0.0 and later · before 3.11.0

Versions reported by the vendor

Affected version ranges reported by the assigning authority (Google). Independent of NVD's CPE analysis and usually ahead of it.

  • Keras-team Keras

    • 3.0.0 and later · before 3.11.0affected · semver

Package-level exposure

OSV and GitHub Advisory data: ecosystem, package and range. SBOM matching uses this table.

EcosystemPackageAffected rangeFix
Debian:11kerasall versions—
PyPIkerasbefore 3.11.03.11.0
PyPIkerasall versions—

Same product

keras: all records

Other highest-scoring records for the same primary product.

  • CVE-2024-3660Arbitrary code injection vulnerability in Keras framework < 2.13
    40Plan
  • CVE-2026-12481Deserialization of Untrusted Data in keras-team/keras
    39Monitor
  • CVE-2025-49655Deserialization of untrusted data can occur in versions of the Keras framework running versions 3.11.0 up to but not including 3.11.3, enabl
    39Monitor
  • CVE-2025-12060Keras keras.utils.get_file Utility Path Traversal Vulnerability
    35Monitor
  • CVE-2025-8747Keras safe_mode bypass allows arbitrary code execution when loading a malicious model.
    34Monitor
  • CVE-2026-11816Path Traversal in keras-team/keras
    32Monitor

Remediation

Which version to upgrade to

Fix versions compiled from the vendor, package registries and Microsoft. Verify the vendor's note before upgrading.

Product / packageFixed versionSource
Keras-team Keras3.11.0Vendor (CNA)
azl3 keras 3.3.3-3 on Azure Linux 3.03.3.3-4 · CBL-Mariner ReleasesMicrosoft (MSRC)
PyPI:keras3.11.0Package registry (OSV)

Exploit status

No known public exploit

No public exploit has been observed yet. That doesn't mean you're safe, only that the bar is a little higher.

Research context

For pentesters and researchers: attack profile, score disagreement, timeline, patch commits, credits, variant and chain candidates, bug bounty scope. All derived from existing data; no exploit code.

Timeline

From publication to today: proof of concept, Metasploit module, CISA KEV and fix record. Dates are as reported by the sources.

No dated events beyond publication.

EPSS, last 120 days

FIRST EPSS daily score; only changes of 0.01 or more are recorded (step chart).

Patch and commit links

Commit, PR and diff links among the references. A starting point for patch-diffing and variant hunting; fixes, not exploits.

Credits

All researchers

Finders, reporters and analysts named in the CNA record. Click a name for that researcher’s other records.

Variant candidates

Same product, same weakness class, within 18 months. If the patch missed the root cause, the sibling bug is here.

  • CVE-2025-4965528 days apartDeserialization of untrusted data can occur in versions of the Keras framework running versions 3.11.0 up to but not including 3.11.3, enabl
    39Monitor
  • CVE-2026-12481287 days apartDeserialization of Untrusted Data in keras-team/keras
    39Monitor
  • CVE-2025-874739 days apartKeras safe_mode bypass allows arbitrary code execution when loading a malicious model.
    34Monitor
  • CVE-2026-1462206 days apartSafe Mode Bypass in keras-team/keras
    31Monitor
  • CVE-2025-1205840 days apartVulnerability in Keras Model.load_model Leading to Arbitrary Local File Loading and SSRF
    23Monitor

Chain candidates

An authentication bypass and a privilege-requiring bug in the same product, published close together: combined they may become an unauthenticated path.

—

Bug bounty scope

No known public program.

Source: bounty-targets-data (public HackerOne, Bugcrowd, Intigriti, YesWeHack listings).

Technical details

Attack conditions

  • Someone with local access to the system can trigger it.
  • A low-privileged account is enough.
  • No special conditions are required; it is repeatable.

If successful

Confidentiality
Integrity
Availability
Attack vector
Local
Attack complexity
Low
Privileges required
Low
User interaction
P
Scope
X

Weakness class (CWE)

CWE-502 · Deserialization of Untrusted Data

CVSS vector

CVSS:4.0/AV:L/AC:L/AT:N/PR:L/UI:P/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:Y/R:A/V:X/RE:X/U:X

nvd-secondary

Root cause

Data received over the network is deserialized into the language’s native object format without verifying its source. Code paths inside the object can be triggered during deserialization.

The bug in code

The whole CWE-502 class

A representative example of this vulnerability class. Not the vendor's source code; it shows the faulty pattern and its fix.

Vulnerable

python
payload = sock.recv(65536)job = pickle.loads(payload)

Fixed

python
payload = sock.recv(65536)if not hmac.compare_digest(sign(payload), header_sig):    raise PermissionError("invalid signature")job = JobSchema.validate(json.loads(payload))

Attack context

MITRE CAPEC attack patterns and ATT&CK techniques for this weakness class (CWE). A starting point for detection rules and threat hunting.

ATT&CK techniques

—

Change log

  1. Fix✗ → ✓

For records you follow, these changes also arrive as notifications. →

References

Vendor advisories and official records. Exploit/PoC links are deliberately left out.

All records