Kaspersky Antimalware ML Model Evaded via Feature Knowledge

# Kaspersky Antimalware ML Model Evaded via Feature Knowledge

https://gcdradar.vercel.app/incidents/kaspersky-antimalware-ml-model-evaded-via-feature-knowledge · AIR-2023-0011 · closed 2023-04-14

- Lane: Research
- Severity: 1 (Research / proof of concept)
- Fix status: Fix status unknown
- First reported: 2023-03-31
- AI role: Attack on AI
- Products: Kaspersky Antimalware ML Model

## Summary

Researchers demonstrated that feature knowledge is sufficient to conduct adversarial attacks against ML-based malware detectors. The technique successfully evaded detection for most modified malware files without white-box access.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2023-03-31: Article, "How to confuse antimalware neural networks. Adversarial attacks and protection" (https://securelist.com/how-to-confuse-antimalware-neural-networks-adversarial-attacks-and-protection/102949)
- 2023-03-31: Confusing Antimalware Neural Networks (https://atlas.mitre.org/studies/AML.CS0014)
- 2023-03-31: AVID-2023-V014: Cloud storage and computations have become popular platforms for deploying ML malware detectors. (https://github.com/avidml/avid-db/blob/main/vulnerabilities/2023/AVID-2023-V014.json)

## Sources

- 2023-03-31 AVID: AVID-2023-V014: Cloud storage and computations have become popular platforms for deploying ML malware detectors. https://github.com/avidml/avid-db/blob/main/vulnerabilities/2023/AVID-2023-V014.json
- 2023-03-31 MITRE ATLAS: Confusing Antimalware Neural Networks https://atlas.mitre.org/studies/AML.CS0014
- 2023-03-31 Kaspersky Securelist: Article, "How to confuse antimalware neural networks. Adversarial attacks and protection" https://securelist.com/how-to-confuse-antimalware-neural-networks-adversarial-attacks-and-protection/102949

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.