SKIDIA's PaperBoy

EN WIRE · 2026-09-28 18:45

Suspicions Mount That AI Services Trained Models on User Conversations Without Consent

Mostly True

Suspicions Mount That AI Services Trained Models on User Conversations Without Consent
Image source: 조사 출처

A series of allegations has been raised that major AI developers used user conversations to train their large language models (LLMs) without obtaining explicit consent, drawing renewed attention to data privacy practices across the AI industry. Primary sources reviewed in the assessment numbered 12.

원문 주장 (KR)
사용자 대화 '무단 학습' 의혹 잇달아 제기

Allegations Center on Consent Gaps

The claims concern the practice of feeding user chat logs into model training pipelines. Critics argue that in cases where users did not opt in to data collection for training purposes, such use constitutes unauthorized learning on personal conversations.

The issue has fueled debate over whether current consent frameworks used by AI service providers adequately inform users of how their dialogue data may be handled, and whether opt-out mechanisms are sufficient.

Privacy Oversight Under Scrutiny

The allegations arrive as regulators and data protection authorities worldwide weigh how existing privacy rules apply to generative AI services. In Korea, the Ministry of Science and ICT and the Personal Information Protection Commission have been focal points for oversight of AI data practices, though the specific regulatory response to these latest suspicions remains to be seen.

Companies named in or connected to the allegations have not uniformly confirmed the practices described, and the extent to which user conversations were used for training without consent remains unclear in some cases.

Uncertainty Remains on Scope

While the core suspicion — that user conversations may have been used for model training without explicit user permission — is supported by the available record, key details such as the scale of the data involved and which specific services were affected could not be fully established.

The underlying claim is judged to be largely accurate, though with gaps in confirmed detail. Verdict: Mostly True

Sources — primary documents (12)
  1. https://www.nytimes.com/2026/09/27/science/anthropic-biology-enzyme-mestre.html
  2. https://www.n
  3. https://www.khan.co.kr/article/202609281715001
  4. https://www.anthropic.com/news/claude-discovers-novel-enzyme-system
  5. https://www-cdn.anthropic.com/22573675ada52a8ca8a97a1a4b4326b2f208a071.pdf
  6. https://openai.com/index/navier-stokes-solution/
  7. https://www.anthropic.com/news/updates-to-our-consumer-terms
  8. https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training
  9. https://www.business-standard.com/technology/artificial-intelligence/did-anthropic-s-ai-really-make-a-scientific-discovery-on-its-own-126092800073_1.html
  10. https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-train-models
  11. https://openaipolicies/terms-of-use
  12. https://openai.com/policies/terms-of-use/

KR: /news/20260928-345429 · 판정: 대체로 사실