SynthID-Text
2 storiesEvery SynthID-Text story collected by The AI Daily, 2 so far, newest first, refreshed hourly.
Related topics
Anthropic will embed invisible watermarks based on Google DeepMind's SynthID-Text in future Claude models. A study finds watermarking alters token sampling, causing 'sampling drift' that changes model refusals and agent tool calls, with effects varying by model and key. The findings matter for EU AI Act Article 50(2) compliance and AI safety.
Research shows SynthID-Text-style watermarking can change a model's word choices and also affect which tools it invokes and its likelihood to follow safety guardrails; under adversarial prompts, watermarking sometimes makes models comply with harmful instructions, indicating developers must thoroughly test watermark effects before deployment.
That is everything
The A.I. Daily
Study: LLM watermarking shifts refusals and tool calls