Researchers replicate machine translation services and transfer adversarial attacks

Google Translate, Bing Translator, Systran Translate

Academic researchers successfully replicated commercial translation APIs and used the extraction to craft adversarial inputs causing targeted translation errors. The work demonstrates model extraction and adversarial vulnerability.

first report: 31 Mar 2023
last update: 31 Mar 2023

Timeline · 5

Sources · 5

News and write-ups

2023-03-31
Project Page, "Imitation Attacks and Defenses for Black-box Machine Translation Systems"
ericswallace.com · ericswallace.com/imitation
2023-03-31
Google under fire for mistranslating Chinese amid Hong Kong protests
thehill.com · thehill.com/policy/international/asia-pacific/449164-google-under-fire-for-mistr

Research and catalogues

2023-03-31
AVID-2023-V006: Machine translation services (such as Google Translate, Bing Translator, and Systran Translate) provide public-facing UIs and APIs.
AVID · github.com/avidml/avid-db/blob/main/vulnerabilities/2023/AVID-2023-V006.json
2023-03-31
Attack on Machine Translation Service - Google Translate, Bing Translator, and Systran Translate
MITRE ATLAS · atlas.mitre.org/studies/AML.CS0005
2023-03-31
Wallace, Eric, et al. "Imitation Attacks and Defenses for Black-box Machine Translation Systems" EMNLP 2020
arXiv · arxiv.org/abs/2004.15015

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: Model theft / extractionAdversarial evasion
severity: 1 · Research / proof of concept
targets: Google, Enterprise deployment
owasp llm top 10: LLM10:2025 Unbounded Consumption
mitre atlas: AML.T0015 Evade AI Model AML.T0043 Craft Adversarial Data AML.T0048.004 External Harms: AI Intellectual Property Theft
tags: model-extraction adversarial-evasion translation
dates: happened 31 Mar 2023 · first reported 31 Mar 2023
origin: AVID
record: confidence 0.95 · 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

# Researchers replicate machine translation services and transfer adversarial attacks

https://gcdradar.vercel.app/incidents/researchers-replicate-machine-translation-services-and-trans · AIR-2023-0008 · 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: Google Translate, Bing Translator, Systran Translate

## Summary

Academic researchers successfully replicated commercial translation APIs and used the extraction to craft adversarial inputs causing targeted translation errors. The work demonstrates model extraction and adversarial vulnerability.

## What to do

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

## Latest updates

- 2023-03-31: Google under fire for mistranslating Chinese amid Hong Kong protests (https://thehill.com/policy/international/asia-pacific/449164-google-under-fire-for-mistranslating-chinese-amid-hong-kong)
- 2023-03-31: Project Page, "Imitation Attacks and Defenses for Black-box Machine Translation Systems" (https://ericswallace.com/imitation)
- 2023-03-31: Wallace, Eric, et al. "Imitation Attacks and Defenses for Black-box Machine Translation Systems" EMNLP 2020 (https://arxiv.org/abs/2004.15015)
- 2023-03-31: Attack on Machine Translation Service - Google Translate, Bing Translator, and Systran Translate (https://atlas.mitre.org/studies/AML.CS0005)
- 2023-03-31: AVID-2023-V006: Machine translation services (such as Google Translate, Bing Translator, and Systran Translate) provide public-facing UIs and APIs. (https://github.com/avidml/avid-db/blob/main/vulnerabilities/2023/AVID-2023-V006.json)

## Sources

- 2023-03-31 AVID: AVID-2023-V006: Machine translation services (such as Google Translate, Bing Translator, and Systran Translate) provide public-facing UIs and APIs. https://github.com/avidml/avid-db/blob/main/vulnerabilities/2023/AVID-2023-V006.json
- 2023-03-31 MITRE ATLAS: Attack on Machine Translation Service - Google Translate, Bing Translator, and Systran Translate https://atlas.mitre.org/studies/AML.CS0005
- 2023-03-31 arXiv: Wallace, Eric, et al. "Imitation Attacks and Defenses for Black-box Machine Translation Systems" EMNLP 2020 https://arxiv.org/abs/2004.15015
- 2023-03-31 ericswallace.com: Project Page, "Imitation Attacks and Defenses for Black-box Machine Translation Systems" https://ericswallace.com/imitation
- 2023-03-31 thehill.com: Google under fire for mistranslating Chinese amid Hong Kong protests https://thehill.com/policy/international/asia-pacific/449164-google-under-fire-for-mistranslating-chinese-amid-hong-kong

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

closed
Commercial Face Identification Service MITRE AI Red Team demonstrates physical evasion attack on face identification
Research · since 31 Mar 2023
AI
closed
Palo Alto Networks DGA Detector CNN-Based Botnet DGA Detector Bypassed via Domain Mutation
Research · since 31 Mar 2023
AI
closed
Kaspersky Antimalware ML Model Kaspersky Antimalware ML Model Evaded via Feature Knowledge
Research · since 31 Mar 2023
AI
closed
Palo Alto Networks C&C Traffic Detector Palo Alto Networks security team demonstrates evasion of deep learning malware detector
Research · since 31 Mar 2023
AI
closed
Shanghai government tax office facial recognition service Camera hijack attack evades facial recognition in tax system fraud
Breaches & attacks · Real loss · since 2018 (approx.)
AI
closed
Hive Box Facial-Recognition Locks Hive Box facial-recognition delivery lockers bypassed with printed photos
Research · since 23 Dec 2022
AI