Azure Red Team demonstrates adversarial evasion on Microsoft edge AI product
Microsoft Edge AI Product
The Azure Red Team conducted an automated red team exercise that continuously manipulated target images to induce misclassifications in a Microsoft edge AI product. The finding highlights adversarial evasion risks in machine learning models.
- first report
- 31 Mar 2023
- last update
- 31 Mar 2023
Timeline · 2
Sources · 2
Research and catalogues
- 2023-03-31AVID-2023-V011: The Azure Red Team performed a red team exercise on a new Microsoft product designed for running AI workloads at the edge.AVID · github.com/avidml/avid-db/blob/main/vulnerabilities/2023/AVID-2023-V011.json
- 2023-03-31Microsoft Edge AI EvasionMITRE ATLAS · atlas.mitre.org/studies/AML.CS0011
Titles and links belong to their publishers. The summary above is written by gcdRadar in its own words; data is CC BY-SA 4.0.
Classification
- lane
- Research
- AI role
- Attack on AI
- categories
- Adversarial evasion
- severity
- 1 · Research / proof of concept
- targets
- Microsoft
- owasp llm top 10
- no entry for these categories
- dates
- happened 31 Mar 2023 · first reported 31 Mar 2023
- origin
- AVID
- record
- confidence 0.85 · created 2026-10-18 06:56Z · updated 2026-10-18 06:56Z
Lane, AI role, categories and severity come from the classifier; the OWASP and ATLAS rows are derived from the categories. How these are assigned
Copy a brief
This record as a short Markdown brief, for a ticket, a chat or a coding agent.
the brief as text
# Azure Red Team demonstrates adversarial evasion on Microsoft edge AI product https://gcdradar.vercel.app/incidents/azure-red-team-demonstrates-adversarial-evasion-on-microsoft · AIR-2023-0017 · 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: Microsoft Edge AI Product ## Summary The Azure Red Team conducted an automated red team exercise that continuously manipulated target images to induce misclassifications in a Microsoft edge AI product. The finding highlights adversarial evasion risks in machine learning models. ## What to do No action recorded; read the sources below and the vendor's advisory. ## Latest updates - 2023-03-31: Microsoft Edge AI Evasion (https://atlas.mitre.org/studies/AML.CS0011) - 2023-03-31: AVID-2023-V011: The Azure Red Team performed a red team exercise on a new Microsoft product designed for running AI workloads at the edge. (https://github.com/avidml/avid-db/blob/main/vulnerabilities/2023/AVID-2023-V011.json) ## Sources - 2023-03-31 AVID: AVID-2023-V011: The Azure Red Team performed a red team exercise on a new Microsoft product designed for running AI workloads at the edge. https://github.com/avidml/avid-db/blob/main/vulnerabilities/2023/AVID-2023-V011.json - 2023-03-31 MITRE ATLAS: Microsoft Edge AI Evasion https://atlas.mitre.org/studies/AML.CS0011 Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.
Related
- closedAzure Microsoft AI Red Team conducts red teaming exercise on Azure service
Research · since 31 Mar 2023
AI - closedCylancePROTECT Researchers bypass Cylance AI malware detector with universal string
Research · since 31 Mar 2023
AI - closedCommercial Face Identification Service MITRE AI Red Team demonstrates physical evasion attack on face identification
Research · since 31 Mar 2023
AI - closedPalo Alto Networks DGA Detector CNN-Based Botnet DGA Detector Bypassed via Domain Mutation
Research · since 31 Mar 2023
AI - closedKaspersky Antimalware ML Model Kaspersky Antimalware ML Model Evaded via Feature Knowledge
Research · since 31 Mar 2023
AI - closedGoogle Translate Researchers replicate machine translation services and transfer adversarial attacks
Research · since 31 Mar 2023
AI