TargetVideo

Tool Removes Claude Watermark Four Hours After Announcement

Following the EU AI labeling rules in effect since early August 2026, Anthropic added an invisible watermark to content generated by its Claude models. Wired reports a workaround appeared only four hours after Anthropic's announcement: developer Guillaume Meyer published a GitHub project called "Watermarks Remover" that aims to remove SynthID-style watermarks by reprocessing text through a non-watermarking model and iteratively replacing words and reordering sentences. The GitHub repo has attracted significant attention (over 15,000 stars and more than 1,700 forks) and multiple contributors. Researchers including Leon Chlon have proposed alternative removal methods such as round-trip translation. Anthropic warns that heavy edits, summaries, or translations can remove its watermark, and the effectiveness of removal tools still needs independent verification.

Google's Gemini Spark AI Agents Arrive, Prompt Engineering Questioned

Google has rolled out its AI agent product Gemini Spark in the U.S. and plans a German release in Q3 2026. A t3n MeisterPrompter podcast with product manager Julia Steier demonstrates everyday use cases—automating email triage, booking restaurants, writing in a matched style—and highlights that users may share private and work data with agents. While agents allow speaking rather than composing ordered prompts, Steier says prompt understanding remains important for complex tasks. The article also raises privacy, data integration, and hallucination concerns around AI agents.

Mistral Shifts Away From European ChatGPT Ambition

The article reports that French AI startup Mistral appears to be moving away from pursuing a frontier, ChatGPT‑style generalist model as a solely European effort. Despite EU state aid and large funding rounds, Mistral has intensified its Microsoft partnership and made GLM‑5.2 (an open‑weights model from China’s Z.ai) available via its infrastructure. Benchmarks from Artificial Analysis show GLM‑5.2 outperforming Mistral’s models, and the company is prioritizing alternative or niche use cases (e.g., OCR) rather than competing head‑on with OpenAI, Anthropic or Meta. The piece frames this as a setback for a Europe‑centric frontier‑model strategy but suggests it could free capacity for specialized, industry‑focused models and niche projects like Soofi S.