📧 info@ciso.sa | 📱 +966550939344 | Riyadh, Kingdom of Saudi Arabia
🔧 Scheduled Maintenance — Saturday 2:00-4:00 AM AST. Some features may be temporarily unavailable.    ●   
💎
Pro Plan 50% Off Unlock all AI features, unlimited reports, and priority support. Upgrade
Search Center
ESC to close
Global general Technology and Artificial Intelligence MEDIUM 1h Global general Technology and Artificial Intelligence HIGH 2h Global vulnerability Higher Education CRITICAL 11h Global data_breach Government HIGH 12h Global supply_chain Software Development and Open Source Communities CRITICAL 12h Global malware Software Development CRITICAL 12h Global phishing Multiple Sectors HIGH 12h Global vulnerability Web Applications CRITICAL 13h Global apt Critical Infrastructure CRITICAL 13h Global ransomware Multiple sectors CRITICAL 14h Global general Technology and Artificial Intelligence MEDIUM 1h Global general Technology and Artificial Intelligence HIGH 2h Global vulnerability Higher Education CRITICAL 11h Global data_breach Government HIGH 12h Global supply_chain Software Development and Open Source Communities CRITICAL 12h Global malware Software Development CRITICAL 12h Global phishing Multiple Sectors HIGH 12h Global vulnerability Web Applications CRITICAL 13h Global apt Critical Infrastructure CRITICAL 13h Global ransomware Multiple sectors CRITICAL 14h Global general Technology and Artificial Intelligence MEDIUM 1h Global general Technology and Artificial Intelligence HIGH 2h Global vulnerability Higher Education CRITICAL 11h Global data_breach Government HIGH 12h Global supply_chain Software Development and Open Source Communities CRITICAL 12h Global malware Software Development CRITICAL 12h Global phishing Multiple Sectors HIGH 12h Global vulnerability Web Applications CRITICAL 13h Global apt Critical Infrastructure CRITICAL 13h Global ransomware Multiple sectors CRITICAL 14h
Vulnerabilities

CVE-2026-0897

High
Allocation of Resources Without Limits or Throttling in the HDF5 weight loading component in Google Keras 3.0.0 through 3.13.0 on all platforms allows a remote attacker to cause a Denial of Service (D
CWE-770 — Weakness Type
Published: Jan 15, 2026  ·  Modified: Feb 28, 2026  ·  Source: NVD
CVSS v3
7.5
🔗 NVD Official
📄 Description (English)

Allocation of Resources Without Limits or Throttling in the HDF5 weight loading component in Google Keras 3.0.0 through 3.13.0 on all platforms allows a remote attacker to cause a Denial of Service (DoS) through memory exhaustion and a crash of the Python interpreter via a crafted .keras archive containing a valid model.weights.h5 file whose dataset declares an extremely large shape.

🤖 AI Executive Summary

CVE-2026-0897 is a Denial of Service vulnerability in Google Keras 3.0.0-3.13.0 affecting HDF5 weight loading. Attackers can craft malicious .keras archive files with oversized dataset declarations to exhaust memory and crash Python interpreters. While no public exploit exists, the vulnerability poses significant risk to organizations using Keras for machine learning model deployment and inference, particularly in cloud and edge computing environments.

📄 Description (Arabic)

🤖 AI Intelligence Analysis Analyzed: May 3, 2026 07:18
🇸🇦 Saudi Arabia Impact Assessment
Saudi organizations in financial technology (fintech), banking AI systems (SAMA-regulated), government AI initiatives (NCA oversight), healthcare AI diagnostics, and energy sector predictive analytics are at risk. ARAMCO's digital transformation initiatives using ML, STC's network optimization systems, and SAMI's defense applications could be impacted. The vulnerability primarily affects development and production environments where Keras models are loaded from untrusted sources or where model repositories lack proper validation controls.
🏢 Affected Saudi Sectors
Banking and Financial Services Government and Defense Healthcare and Medical AI Energy and Utilities Telecommunications Technology and Software Development Research and Academia
⚖️ Saudi Risk Score (AI)
6.8
/ 10.0
🔧 Remediation Steps (English)
1. IMMEDIATE ACTIONS:
- Identify all systems running Keras 3.0.0-3.13.0 across development, testing, and production environments
- Restrict loading of .keras files from untrusted sources until patching is complete
- Implement file integrity verification for model archives before loading

2. PATCHING GUIDANCE:
- Upgrade to Keras 3.13.1 or later immediately
- For TensorFlow users: update to TensorFlow 2.16.0+ which includes patched Keras
- Test patches in non-production environments first

3. COMPENSATING CONTROLS (if immediate patching not possible):
- Implement resource limits: set Python process memory limits using ulimit or container memory constraints
- Deploy model loading in isolated containers with memory caps (e.g., Docker memory limits)
- Validate .keras archive structure before loading using external tools
- Implement timeout mechanisms for model loading operations
- Use sandboxed environments for untrusted model evaluation

4. DETECTION RULES:
- Monitor for Python processes with abnormal memory growth during model loading
- Alert on .keras file loads with dataset shape declarations exceeding 1GB equivalent
- Track failed model loading attempts with memory exhaustion errors
- Log all model file sources and validate against approved repositories
🔧 خطوات المعالجة (العربية)
1. الإجراءات الفورية:
- تحديد جميع الأنظمة التي تعمل بـ Keras 3.0.0-3.13.0 في بيئات التطوير والاختبار والإنتاج
- تقييد تحميل ملفات .keras من مصادر غير موثوقة حتى اكتمال التصحيح
- تنفيذ التحقق من سلامة الملفات لأرشيفات النماذج قبل التحميل

2. إرشادات التصحيح:
- الترقية إلى Keras 3.13.1 أو أحدث فوراً
- لمستخدمي TensorFlow: تحديث إلى TensorFlow 2.16.0+ الذي يتضمن Keras المصحح
- اختبار التصحيحات في بيئات غير الإنتاج أولاً

3. الضوابط البديلة (إذا لم يكن التصحيح الفوري ممكناً):
- تنفيذ حدود الموارد: تعيين حدود ذاكرة عملية Python باستخدام ulimit أو قيود ذاكرة الحاوية
- نشر تحميل النموذج في حاويات معزولة بحدود ذاكرة (مثل حدود ذاكرة Docker)
- التحقق من بنية أرشيف .keras قبل التحميل باستخدام أدوات خارجية
- تنفيذ آليات المهلة الزمنية لعمليات تحميل النموذج
- استخدام بيئات معزولة لتقييم النماذج غير الموثوقة

4. قواعد الكشف:
- مراقبة عمليات Python بنمو ذاكرة غير طبيعي أثناء تحميل النموذج
- التنبيه على تحميل ملفات .keras بتصريحات شكل مجموعة بيانات تتجاوز 1GB
- تتبع محاولات تحميل النموذج الفاشلة مع أخطاء استنزاف الذاكرة
- تسجيل جميع مصادر ملفات النموذج والتحقق من المستودعات المعتمدة
📋 Regulatory Compliance Mapping
🟢 NCA ECC 2024
A.5.1.1 - Information security policies and procedures A.8.1.1 - User access management A.12.2.1 - Change management procedures A.12.6.1 - Management of technical vulnerabilities
🔵 SAMA CSF
ID.RA-1 - Asset management and criticality assessment PR.IP-1 - Information and records management PR.IP-12 - Software development and quality assurance DE.CM-8 - Vulnerability scans
🟡 ISO 27001:2022
A.12.2.1 - Change management A.12.6.1 - Management of technical vulnerabilities A.14.2.1 - Secure development policy A.14.2.5 - Secure development environment
🟣 PCI DSS v4.0.1
6.2 - Security patches and updates 6.3.1 - Vulnerability identification and remediation
📦 Affected Products / CPE 1 entries
keras:keras
📊 CVSS Score
7.5
/ 10.0 — High
📊 CVSS Vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H
Attack VectorN — None / Network
Attack ComplexityL — Low / Local
Privileges RequiredN — None / Network
User InteractionN — None / Network
ScopeU — Unchanged
ConfidentialityN — None / Network
IntegrityN — None / Network
AvailabilityH — High
📋 Quick Facts
Severity High
CVSS Score7.5
CWECWE-770
EPSS0.03%
Exploit No
Patch ✓ Yes
Published 2026-01-15
Source Feed nvd
Views 5
🇸🇦 Saudi Risk Score
6.8
/ 10.0 — Saudi Risk
Priority: HIGH
🏷️ Tags
patch-available CWE-770
Share this CVE

💬 Comments

0
Loading comments
📣 Found this valuable?
Share it with your cybersecurity network
in LinkedIn 𝕏 X / Twitter 💬 WhatsApp ✈ Telegram
🍪 Privacy Preferences
CISO Consulting — Compliant with Saudi Personal Data Protection Law (PDPL)
We use cookies and similar technologies to provide the best experience on our platform. You can choose which types you accept.
🔒
Essential Always On
Required for the website to function properly. Cannot be disabled.
📋 Sessions, CSRF tokens, authentication, language preferences
📊
Analytics
Help us understand how visitors use the site and improve performance.
📋 Page views, session duration, traffic sources, performance metrics
⚙️
Functional
Enable enhanced features like content personalization and preferences.
📋 Dark/light theme, font size, custom dashboards, saved filters
📣
Marketing
Used to deliver content and ads relevant to your interests.
📋 Campaign tracking, retargeting, social media analytics
Privacy Policy →
CISO AI Assistant
Ask anything · Documents · Support
🔐

Introduce Yourself

Enter your details to access the full assistant

Your info is private and never shared
💬
CyberAssist
Online · responds in seconds
5 / 5
🔐 Verify Your Identity

Enter your email to receive a verification code before submitting a support request.

Enter to send · / for commands 0 / 2000
CISO AI · Powered by Anthropic Claude
✦ Quick Survey Help Us Improve CISO Consulting Your feedback shapes the future of our platform — takes less than 2 minutes.
⚠ Please answer this question to continue

How would you rate your overall experience with our platform?

Rate from 1 (poor) to 5 (excellent)

🎉
Thank you!
Your response has been recorded.