VulnerabilityPublished 11 May 2026Updated 12 May 2026

CVE-2026-31253: flash-attention training framework thru · Deserialization of Untrusted Data

Source CVE Database V5

Not known to be exploited · CVSS 7.3 High.

Data as of 26 Oct 2026, 17:16 UTC

CVSS 3.1 base score

7.3High

How severe the flaw is on the 0–10 CVSS scale.

EPSS: 0.2% chance of exploitation in 30 days – higher than 12% of CVEsExploit Prediction Scoring System: the likelihood attackers will use this flaw soon.

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

Attack Vector: Network
Attack Complexity: Low
Privileges Required: None
User Interaction: None
Scope: Unchanged
Confidentiality: Low
Integrity: Low
Availability: Low

Bold values are the worst case: easiest to exploit or largest impact.

Attack Vector: Network: The vulnerable component is bound to the network stack and the set of possible attackers extends beyond the other options listed, up to and including the entire Internet. Such a vulnerability is often termed 'remotely exploitable' and can be thought of as an attack being exploitable at the protocol level one or more network hops away (e.g., across one or more routers).
Attack Complexity: Low: Specialized access conditions or extenuating circumstances do not exist. An attacker can expect repeatable success against the vulnerable component.
Privileges Required: None: The attacker is unauthorized prior to attack, and therefore does not require any access to settings or files to carry out an attack.
User Interaction: None: The vulnerable system can be exploited without any interaction from any user.
Scope: Unchanged: An exploited vulnerability can only affect resources managed by the same security authority. In this case, the vulnerable component and the impacted component are either the same, or both are managed by the same security authority.
Confidentiality: Low: There is some loss of confidentiality. Access to some restricted information is obtained, but the attacker does not have control over what information is obtained, or the amount or kind of loss is limited. The information disclosure does not cause a direct, serious loss to the impacted component.
Integrity: Low: Modification of data is possible, but the attacker does not have control over the consequence of a modification, or the amount of modification is limited. The data modification does not have a direct, serious impact on the impacted component.
Availability: Low: Performance is reduced or there are interruptions in resource availability. Even if repeated exploitation of the vulnerability is possible, the attacker does not have the ability to completely deny service to legitimate users. The resources in the impacted component are either partially available all of the time, or fully available only some of the time, but overall there is no direct, serious consequence to the impacted component.

Definitions: FIRST CVSS v3.1 specification.

Exploitation evidence

Disclosed: evidence found.
CVE published 11 May 2026
Detectable: no evidence on record.
No public nuclei template indexed
Weaponised: no evidence on record.
No public exploit code indexed in Metasploit or Exploit-DB
Exploited in the wildExploited: no evidence on record.
Not in CISA KEV and no source reports exploitation
Observed by OffSeq MirageSeen by Mirage: no evidence on record.
No attack traffic recorded by OffSeq Mirage sensors

A missing step is not evidence of safety. Sources: cve.org (opens in a new tab), the Metasploit, Exploit-DB and nuclei-templates index as of 26 Oct 2026.

Disclosed: The CVE record is public on cve.org. Its publication is t = 0.
Detectable: A public nuclei template can find exposed instances, for defenders and attackers alike.
Weaponised: Public exploit code exists in Metasploit or Exploit-DB, or CISA SSVC or the CVSS exploit maturity reports a public proof of concept.
Exploited in the wild: CISA KEV lists it, CISA SSVC rates exploitation active, or a source reports attacks.
Observed by OffSeq Mirage: OffSeq Mirage honeypot sensors recorded attack traffic for this CVE, or found exposed hosts without attack traffic.

Moments are spaced evenly, not to scale; a dashed line marks a gap of six months or more. Deltas count from t = 0. Sources that give only a date (CISA KEV, Exploit-DB, or a time of exactly 00:00Z) read “same day” rather than hours. Metasploit modules carry the flaw’s disclosure date, not their release date, and nuclei templates carry no date, so both appear above but not on the timeline.

What to do

AI-suggested

Until an official fix is available, users should avoid loading checkpoint files from untrusted or unauthenticated sources. Applying the weights_only=True parameter in torch.load() when loading checkpoints is a recommended security practice to prevent arbitrary code execution via deserialization.

Description

The flash-attention training framework thru commit e724e2588cbe754beb97cf7c011b5e7e34119e62 (2025-13-04) contains an insecure deserialization vulnerability (CWE-502) in its checkpoint loading mechanism. The load_checkpoint() function in checkpoint.py and the checkpoint loading code in eval.py use torch.load() without enabling the security-restrictive weights_only=True parameter. This allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing a maliciously crafted checkpoint file. When a victim loads this checkpoint during model warmstarting or evaluation, arbitrary code is executed on the victim's system.

Impact

An attacker who can supply a malicious checkpoint file to a victim using the vulnerable flash-attention training framework can achieve arbitrary code execution on the victim's system during checkpoint loading. This compromises the confidentiality, integrity, and availability of the affected system. There are no known exploits in the wild at this time.

Summary

The flash-attention training framework contains an insecure deserialization vulnerability (CWE-502) due to unsafe use of torch.load() in checkpoint.py and eval.py. Specifically, the load_checkpoint() function loads checkpoint files without enabling the weights_only=True parameter, which would restrict deserialization to tensor weights only. This improper use allows arbitrary Python objects to be deserialized via the pickle module, enabling remote code execution if a malicious checkpoint file is loaded. The vulnerability is identified as CVE-2026-31253 with a CVSS 3.1 base score of 7.3 (AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:L/A:L), indicating network attack vector, low attack complexity, no privileges or user interaction required, and impacts on confidentiality, integrity, and availability.

CWE-502Deserialization of Untrusted Data (opens in a new tab)
Base weaknessLikelihood of exploit: MediumCWE Top 25 (2025): #15OWASP Top 10 A08:2025: Software or Data Integrity Failures (opens in a new tab)
The product deserializes untrusted data without sufficiently ensuring that the resulting data will be valid.
Architecture and Design · Implementation
If available, use the signing/sealing features of the programming language to assure that deserialized data has not been tainted. For example, a hash-based message authentication code (HMAC) could be used to ensure that data has not been modified.
Implementation
When deserializing data, populate a new object rather than just deserializing. The result is that the data flows through safe input validation and that the functions are safe.
Implementation
Explicitly define a final object() to prevent deserialization.
OWASP Cheat Sheets
Abuse Case (opens in a new tab)
Deserialization (opens in a new tab)
XML Security (opens in a new tab)
XML External Entity Prevention (opens in a new tab)
Software Supply Chain Security (opens in a new tab)
OWASP ASVS 4.0.3
V1.5.2 (opens in a new tab)
V5.5.3 (opens in a new tab)
V5.5.1 (opens in a new tab)
Attack patterns
Object Injection (opens in a new tab)
From MITRE CWE, MITRE CAPEC and OWASP (Top 10, OpenCRE).
CWE-94Improper Control of Generation of Code ('Code Injection') (opens in a new tab)
Base weaknessLikelihood of exploit: MediumCWE Top 25 (2025): #10OWASP Top 10 A05:2025: Injection (opens in a new tab)
The product constructs all or part of a code segment using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the syntax or behavior of the intended code segment.
Architecture and Design · Refactoring
Refactor your program so that you do not have to dynamically generate code.
Architecture and Design
Run your code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict which code can be executed by your product.
Implementation · Input Validation
Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does.
OWASP Cheat Sheets
Server Side Request Forgery Prevention (opens in a new tab)
Unvalidated Redirects and Forwards (opens in a new tab)
Cross Site Scripting Prevention (opens in a new tab)
DOM based XSS Prevention (opens in a new tab)
Abuse Case (opens in a new tab)
OWASP ASVS 4.0.3
V5.2.8 (opens in a new tab)
V5.2.5 (opens in a new tab)
V1.5.1 (opens in a new tab)
Attack patterns
Leverage Executable Code in Non-Executable Files (opens in a new tab)
Manipulating User-Controlled Variables (opens in a new tab)
Code Injection (opens in a new tab)
From MITRE CWE, MITRE CAPEC and OWASP (Top 10, OpenCRE).

References

Community reviews

Loading reviews…

Exploitation evidence

What to do

Description

AI analysisMachine-generated, updated 19 May 2026, 10:28 UTC

WeaknessesCWE-502, CWE-94

Technical details

References

Community reviews

Related threats