CVE-2025-9906 & CVE-2025-9905 — Keras Library Vulnerabilities Arbitrary Code Execution in AI/ML Framework Vulnerability Analysis Report — By CyberDudeBivash

-



 Author: CyberDudeBivash · Powered by: CyberDudeBivash

Executive Summary

Two serious vulnerabilities were disclosed in the Keras library, widely used in deep learning workflows.

  • CVE-2025-9906: CVSS 8.6 (High)

  • CVE-2025-9905: CVSS 7.3 (High)

Both issues allow arbitrary code execution (ACE), which could be weaponized in supply-chain attacks, malicious model distribution, or unsafe deserialization of model files. Since Keras underpins many AI/ML production pipelines, the impact radius is vast — from research environments to enterprise ML deployments.

Technical Details

CVE-2025-9906 (CVSS 8.6)

  • Type: Deserialization / unsafe model parsing flaw.

  • Impact: Maliciously crafted model files (.h5 / TensorFlow SavedModel) can trigger execution of arbitrary code when loaded.

  • Attack Scenario: An attacker uploads or distributes a tainted model (e.g., via GitHub, Hugging Face, PyPI) → victim loads it into Keras → embedded payload executes.

  • Severity Justification: High (8.6) because exploitation requires crafted input but leads to full compromise of ML host.

CVE-2025-9905 (CVSS 7.3)

  • Type: Input validation flaw in preprocessing utilities.

  • Impact: Under certain conditions, hostile inputs (images, JSON configs, or serialized weight files) cause Keras functions to execute unintended code paths.

  • Attack Scenario: Malicious dataset/model metadata used in pipelines (e.g., CI/CD for ML ops) → triggers RCE during training or inference setup.

  • Severity Justification: Medium-High (7.3) — requires malicious input file or supply-chain poisoning.

Exploitation Risks

  • Supply Chain Poisoning: Malicious models on public repositories can infect enterprise environments.

  • CI/CD Attack Surface: Automated retraining workflows that pull community models are especially at risk.

  • Cloud ML Platforms: Shared GPU/TPU environments may be abused as stepping stones for lateral movement.

  • Data Exfiltration: Attackers can run arbitrary Python code to harvest credentials, data, or inject persistence.

Detection & Indicators

  • Unexpected system calls during keras.models.load_model() execution.

  • Presence of pickled objects / suspicious lambdas in model files.

  • ML pipelines spawning child processes not normally used by training jobs.

  • Integrity mismatches in downloaded models (hash checks failing).

Immediate Mitigations

  1. Upgrade to the patched version of Keras (check PyPI / GitHub releases).

  2. Verify model integrity — only load models from trusted sources; validate SHA256 hashes.

  3. Sandbox risky operations — run model ingestion in restricted containers.

  4. Disable auto-execution features — avoid eval() or pickle-based deserialization in untrusted contexts.

  5. Code reviews — audit ML pipeline code for unsafe load practices.

Longer-Term Recommendations

  • Secure MLOps: Enforce model signing and verification (e.g., Sigstore, cosign).

  • Policy Enforcement: Treat ML artifacts like binaries — with vulnerability scanning and provenance checks.

  • Zero-Trust ML: Assume third-party datasets/models are malicious until validated.

  • Continuous Threat Hunting: Monitor ML workloads for anomalies in system resource usage.

CyberDudeBivash Action Checklist

  •  Patch Keras to latest release across all environments.

  •  Audit ML repos for untrusted .h5 / SavedModel files.

  •  Enforce SHA256 signature verification for every model load.

  •  Run risky ML jobs in containerized sandboxes with restricted privileges.

  •  Monitor for suspicious process execution from ML training pipelines.

  •  Educate data scientists & engineers about malicious model supply-chain threats.

Conclusion

CVE-2025-9906 and CVE-2025-9905 highlight how AI/ML frameworks are becoming prime cyber targets. Exploiting Keras vulnerabilities offers attackers direct execution inside GPU/TPU-equipped environments, often with privileged access. Organizations must patch quickly, enforce model provenance controls, and integrate security into MLOps pipelines.


Affiliate Toolbox (clearly disclosed)

Disclosure: If you buy via the links below, we may earn a commission at no extra cost to you. These items supplement (not replace) your security controls. This supports CyberDudeBivash in creating free cybersecurity content.

🌐 cyberdudebivash.com | cyberbivash.blogspot.com

#CyberDudeBivash #CVE20259906 #CVE20259905 #Keras #MachineLearning #MLOps #AIsecurity #SupplyChainAttack #ThreatIntel #Infosec

Comments

Post a Comment

Popular posts from this blog

CyberDudeBivash Rapid Advisory — WordPress Plugin: Social-Login Authentication Bypass (Threat Summary & Emergency Playbook)

-
Image
  TL;DR: A class of vulnerabilities in WordPress social-login / OAuth plugins can let attackers bypass normal authentication flows and obtain an administrative session (or create admin users) by manipulating OAuth callback parameters, reusing stale tokens, or exploiting improper validation of the identity assertions returned by providers. If you run a site that accepts social logins (Google, Facebook, Apple, GitHub, etc.), treat this as high priority : audit, patch, or temporarily disable social login until you confirm your plugin is safe. This advisory gives you immediate actions, detection steps, mitigation, and recovery guidance. Why this matters (short) Social-login plugins often accept externally-issued assertions (OAuth ID tokens, authorization codes, user info). If the plugin fails to validate provider signatures, nonce/state values, redirect URIs, or maps identities to local accounts incorrectly , attackers can craft requests that the site accepts as authenticated. ...
Read more

Hackers Injecting Malicious Code into GitHub Actions to Steal PyPI Tokens CyberDudeBivash — Threat Brief & Defensive Playbook

-
Image
  SUMMARY  Attackers are modifying GitHub Actions workflows (via compromised accounts, malicious PRs, or forged commits) to add steps that exfiltrate PyPI (or other package registry) publishing tokens. With those tokens they publish backdoored packages or download private packages — a supply-chain multiplier. Defend by eliminating long-lived secrets, switching to OIDC/ephemeral credentials, hardening workflow controls, monitoring workflow changes and secret usage, and rotating tokens immediately if you suspect compromise. How attackers typically operate  Gain write access to repo or CI/CD config (via compromised developer account, stolen token, privilege abuse, or merge of malicious PR). Modify or add a workflow file that runs in the repo’s runner context (self-hosted or GitHub-hosted). Add steps that read repository secrets (e.g., PYPI_TOKEN ) or create credentials files ( ~/.pypirc ), then send the token to attacker-controlled endpoints or push a malicious p...
Read more

Exchange Hybrid Warning: CVE-2025-53786 can cascade into domain compromise (on-prem ↔ M365) By CyberDudeBivash — Cybersecurity & AI

-
Image
  Executive summary CVE-2025-53786 is a post-authentication elevation of privilege in Microsoft Exchange hybrid deployments . If an attacker first gains admin on an on-prem Exchange server , they can abuse the legacy “shared service principal” trust used by Hybrid Configuration to escalate into Exchange Online —potentially leading to full cloud+on-prem domain compromise . CISA issued Emergency Directive ED-25-02 with a deadline of 9:00 a.m. EDT on August 11, 2025 for U.S. federal agencies; Microsoft urges all hybrid customers to take the same steps: apply the April 2025 Hotfix Updates , migrate to the dedicated Exchange Hybrid app , and reset/clean up service principal credentials . cisa.gov +1 TECHCOMMUNITY.MICROSOFT.COM +1 Why this is dangerous: actions initiated from the on-prem Exchange side may not show up as obviously malicious in M365 audit trails, making cloud abuse harder to spot if you’ve not modernized the trust. TechRadar What is actually vulnerable? (trust...
Read more