CVE-2025-9905 | Arbitary Code execution in Keras load_model()

Exp

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 .h5/.hdf5 model archive that, when loaded via Model.load_model, will trigger arbitrary code to be executed. This is achieved by crafting a special .h5 archive file that uses the Lambda layer feature of keras which allows arbitrary Python code in the form of pickled code. The vulnerability comes from the fact that the safe_mode=True option is not honored when reading .h5 archives. Note that the .h5/.hdf5 format is a legacy format supported by Keras 3 for backwards compatibility.

Published: 2025-09-19 Last update: 2025-09-23 Assigner: [email protected] Source: [email protected]
NVD Status：Analyzed，CVE State: published
View at NVD, CVE.org, EUVD

CVE-2025-9905 is rated Exploit Available (50/100): CVSS High severity, with low exploitation likelihood (EPSS 0.21%). 1 public exploit reference(s) are indexed (Exploit-DB). Public exploits are available—assess exposure, apply mitigations, and prioritize patching.

Risk is dynamic; we continuously reassess and refresh what is shown on this page as upstream context changes.

Public exploit references (Exploit-DB) for CVE-2025-9905

EDB-ID Source Kind Published Link
— nvd_ref exploit_tag Exploit-DB ↗

Exploit prediction scoring system (EPSS) score for CVE-2025-9905

EPSS lead: Daily EPSS estimates relative likelihood of exploitation; percentile ranks this CVE among scored vulnerabilities (higher = more severe relative rank).

# Date Old EPSS score New EPSS score Delta (New - Old)
1 2026-06-15 0.01% 0.21% +0.19%
2 2025-09-19 — 0.01% —

Full EPSS history (2 records total)

Common vulnerability scoring system (CVSS) metrics for CVE-2025-9905

CVSS metrics for this CVE.

Base score Version Severity Vector Exploitability Impact Score source
7.3 4.0 HIGH
CVSS:4.0/AV:L/AC:H/AT:P/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:X/R:X/V:X/RE:X/U:X Click to expand
Attack vector (AV:L)
Attacker needs local access on the target system.
Attack complexity (AC:H)
Exploitation depends on constrained or hard-to-reproduce conditions.
Attack requirements (AT:P)
Additional preconditions must be present for exploitation.
Privileges required (PR:L)
Low privileges are required.
User interaction (UI:P)
A user has to participate (for example click/open/approve).
Vulnerable system confidentiality impact (VC:H)
High confidentiality impact on the vulnerable system.
Vulnerable system integrity impact (VI:H)
High integrity impact on the vulnerable system.
Vulnerable system availability impact (VA:H)
High availability impact on the vulnerable system.
Subsequent system confidentiality impact (SC:H)
High confidentiality impact on subsequent systems.
Subsequent system integrity impact (SI:H)
High integrity impact on subsequent systems.
Subsequent system availability impact (SA:H)
High availability impact on subsequent systems.
Exploit maturity (threat) (E:X)
Not defined: no reliable threat intelligence; scoring assumes the worst case (equivalent to Attacked).
Confidentiality requirement (CR:X)
Not defined: insufficient information; scoring treats this like High (worst case).
Integrity requirement (IR:X)
Not defined: insufficient information; scoring treats this like High (worst case).
Availability requirement (AR:X)
Not defined: insufficient information; scoring treats this like High (worst case).
Modified attack vector (MAV:X)
Not defined: scoring uses the Base Attack Vector (AV).
Modified attack complexity (MAC:X)
Not defined: scoring uses the Base Attack Complexity (AC).
Modified attack requirements (MAT:X)
Not defined: scoring uses the Base Attack Requirements (AT).
Modified privileges required (MPR:X)
Not defined: scoring uses the Base Privileges Required (PR).
Modified user interaction (MUI:X)
Not defined: scoring uses the Base User Interaction (UI).
Modified vulnerable system confidentiality impact (MVC:X)
Not defined: scoring uses the Base VC metric.
Modified vulnerable system integrity impact (MVI:X)
Not defined: scoring uses the Base VI metric.
Modified vulnerable system availability impact (MVA:X)
Not defined: scoring uses the Base VA metric.
Modified subsequent system confidentiality impact (MSC:X)
Not defined: scoring uses the Base SC metric.
Modified subsequent system integrity impact (MSI:X)
Not defined: scoring uses the Base SI metric.
Modified subsequent system availability impact (MSA:X)
Not defined: scoring uses the Base SA metric.
Safety (supplemental) (S:X)
Not evaluated.
Automatable (supplemental) (AU:X)
Not evaluated.
Recovery (supplemental) (R:X)
Not evaluated.
Value density (supplemental) (V:X)
Not evaluated.
Vulnerability response effort (supplemental) (RE:X)
Not evaluated.
Provider urgency (supplemental) (U:X)
Not evaluated.
 — — [email protected]
7.3 3.1 HIGH
CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:U/C:H/I:H/A:H Click to expand
Attack vector (AV:L)
They already need access on the box, or another person has to do something wrong; it’s not a remote drive-by.
Attack complexity (AC:L)
Once they can reach the bug, pulling it off is straightforward—no weird race conditions or rare setup.
Privileges required (PR:L)
A normal user session is enough; they don’t have to be admin.
User interaction (UI:R)
A real person has to do something—click, install, enable—otherwise it doesn’t land.
Scope (S:U)
Damage stays in the same “trust bubble” as the broken component—no big spill into unrelated systems.
Confidentiality (C:H)
Serious risk that confidential data gets exposed in a big way.
Integrity (I:H)
They could widely tamper with or forge data—trust in the data is badly hurt.
Availability (A:H)
Could take the service down hard or make it unusable for people who depend on it.
 1.3 5.9 [email protected]

Weakness enumeration for CVE-2025-9905

GitHub Security Advisory for CVE-2025-9905

GHSA-36rr-ww3j-vrjv · Severity: high · Ecosystem: pip — The Keras `Model.load_model` method **silently** ignores `safe_mode=True` and allows arbitrary code execution when a `.h5`/`.hdf5` file is loaded.

OS Trackers for CVE-2025-9905

vendor priority summary link
debian not yet assigned CVE-2025-9905 not yet assigned priority: Debian including 1 source packages (keras), 1 status rows across 1 suites (bullseye): open 1. https://security-tracker.debian.org/tracker/CVE-2025-9905
redhat high — https://access.redhat.com/security/cve/CVE-2025-9905
ubuntu medium CVE-2025-9905 medium priority: Ubuntu including 1 source packages (keras), 7 status rows across 7 suites (bionic, focal, jammy, noble, plucky, questing, upstream): DNE 4, needs-triage 3. https://ubuntu.com/security/CVE-2025-9905

Affected software / configurations for CVE-2025-9905

Vendor Product Version Raw CPE
keras keras >= 3.0.0, < 3.11.3 cpe:2.3:a:keras:keras:*:*:*:*:*:*:*:*

References for CVE-2025-9905

cvelogic Threat Intelligence