Malware & RansomwareHIGH

Malware Detectors Stumble When Evaluated on Different Datasets

Featured image for Malware Detectors Stumble When Evaluated on Different Datasets
HNHelp Net Security·Reporting by Anamarija Pogorelec
Summary by CyberPings Editorial·AI-assisted·Reviewed by Rohit Rana
Ingested:
🎯

Basically, malware detectors often fail when faced with new types of malware they weren't trained on.

Quick Summary

A new study reveals that malware detection models often fail when faced with different types of malware. This gap in effectiveness poses risks for organizations relying on these models. Understanding this issue is crucial for improving endpoint security and adapting to evolving threats.

What Happened

Recent research from the Polytechnic of Porto highlights a critical issue with malware detection models. These models are often trained on specific datasets and evaluated on similar data. However, the malware that enterprises encounter can be vastly different. This discrepancy raises concerns about the effectiveness of static detectors, which are commonly used as a first line of defense against malware attacks.

The European Union Agency for Cybersecurity has identified public administration as the most frequently targeted sector by malware. Ransomware and data intrusions are the primary vectors of these attacks. Many tools used in these intrusions employ obfuscation techniques to evade detection. The study aimed to test how well current machine learning-based static detectors perform when faced with malware that does not match their training data.

Who's Being Targeted

The research evaluated detection pipelines using a standardized feature format across six public Windows PE datasets. Two configurations were tested: one utilized a combination of the EMBER and BODMAS datasets, while the other included ERMDS, designed to challenge detectors with obfuscated samples. The models were assessed not only on their training data but also on external datasets like TRITIUM and INFERNO, which contain naturally occurring threat samples and red team malware, respectively.

The results revealed that while models performed well on their training distribution, their effectiveness dropped significantly when tested on external datasets. This finding is crucial for organizations that rely on these models for endpoint security, as it indicates that their defenses may not be as robust as they appear.

Signs of Infection

When evaluated on their training data, the best-performing models achieved high AUC and F1 scores, indicating strong detection capabilities. However, performance varied drastically on external datasets. For instance, while the models transferred well to TRITIUM, their detection rates dropped considerably on INFERNO and SOREL-20M, the latter being the largest and most temporally diverse dataset. Some configurations even fell to levels where their practical utility was limited.

The study also highlighted the obfuscation problem. Attempts to improve detection on obfuscated samples within the ERMDS dataset inadvertently reduced the models' generalization capabilities on broader datasets. This suggests that training to recognize one type of malware can create blind spots for others, complicating the detection landscape.

How to Protect Yourself

Static detectors are appealing due to their computational efficiency and quick verdicts. However, this study underscores a significant limitation: the benchmark performance of a detector is only relevant if the test data reflects the actual threat landscape. Organizations must consider the diversity of malware they might encounter, including packed malware and temporally shifted samples, which can degrade model performance.

The researchers plan to extend their evaluation to deep learning architectures, focusing on how training data composition impacts detection capabilities at low false positive rates. For organizations, this means continuously reassessing their malware detection strategies and ensuring they are equipped to handle the evolving threat landscape effectively.

🔒 Pro insight: Organizations must diversify their training datasets to enhance malware detection capabilities against evolving threats and obfuscation techniques.

Original article from

HNHelp Net Security· Anamarija Pogorelec
Read Full Article

Related Pings

HIGHMalware & Ransomware

Malware Newsletter Round 91 - Latest Threats and Insights

The latest malware newsletter reveals new threats like Infiniti Stealer and npm supply chain attacks. Developers and organizations must stay alert to evolving risks in cybersecurity.

Security Affairs·
HIGHMalware & Ransomware

Malicious Email Delivers CMD Malware - Privilege Escalation Alert

A malicious email has delivered a .cmd malware file that escalates privileges and bypasses antivirus systems. Users are at risk of significant system compromise. Awareness and immediate action are vital to mitigate this threat.

Security Affairs·
HIGHMalware & Ransomware

Axios NPM Package Compromised - Supply Chain Attack Exposed

A major supply chain attack compromised the Axios NPM package, affecting millions of users. Malicious versions deployed a RAT, posing serious security risks. Swift action was taken to remove the threats.

Trend Micro Research·
HIGHMalware & Ransomware

Brokk Hacked - Play Ransomware Exposes Sensitive Data

Brokk has reportedly been hacked by Play ransomware, leading to the leak of sensitive corporate data. This incident could severely impact the company's reputation and security. Organizations must bolster their defenses to prevent similar breaches.

SC Media·
HIGHMalware & Ransomware

Chaos Malware - New Targeting of 64-bit Linux Servers

Chaos malware has evolved to target 64-bit Linux servers, expanding its attack surface. This shift raises alarms for organizations relying on these systems. Enhanced security measures are now crucial to protect against potential larger-scale attacks.

SC Media·
HIGHMalware & Ransomware

Phorpiex Botnet - Spreading Ransomware and Sextortion Tactics

The notorious Phorpiex botnet is back, spreading ransomware and sextortion schemes. Millions are at risk as it targets users globally. Stay alert and protect your devices from this evolving threat.

Cyber Security News·