AI Safety Controversies Expand: From Grok Abuse to Claude Watermarks and Twitch Data Use
AI 안전 | Sun Aug 16 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 3 sources
Generative AI abuse lawsuits, content identification technology adoption, and user data training controversies are emerging simultaneously.
Analysis
[xAI Grok] faced expanded lawsuit over child sexual abuse material generation [1]
- Victim whose childhood photos at age 11 were altered into over 7
- 000 sexual images joined the lawsuit
- Joined as Jane Doe 4 in the existing lawsuit filed by three Tennessee teenagers
- Raised issues over inadequate measures to prevent explicit image generation targeting minors
- Class action status application in progress
[Anthropic Claude] disclosed how AI-generated text watermarking works [2]
- Measure to comply with the EU AI Act Transparency Code
- Adopted Google DeepMind's SynthID-Text approach
- Embeds markers through patterns of choosing similar words like 'overcast' and 'grey'
- No impact on output quality
- with a separate detection API to be released
- Difficult to remove through light editing but eliminated with complete rewriting
[Amazon Twitch] introduced default setting and opt-out option for using streamer content in AI training [3]
- Generative AI Training can be disabled in the Security and Privacy menu of account settings
- Over 16
- 000 creators expressed opposition to the default-enabled approach
- Twitch product head Mike Minton stated that without default opt-in
- 'no one would participate'
- Related clauses included in the terms of service enacted since March 2024
- It remains unclear since when content has been used for training