Anthropic AI Updates: August 12, 2026

1. Anthropic Says It Will Watermark Text Generated by Claude

Anthropic. Anthropic announced it will apply watermarks to text generated by its Claude models worldwide, with marks the company says may persist through some editing. The goal is to improve provenance and detection of AI-generated content as regulatory and trust pressures rise. Text watermarking is harder than image watermarking because there is less signal to hide in, making the durability claims a notable technical bet for practitioners who need to track content origin. Source

2. Compliance API Now Covers Claude Cowork and Claude Code

Anthropic. Anthropic’s Compliance API now covers Claude Cowork and Claude Code, available in beta for Claude Enterprise customers. The expanded endpoints deliver consolidated session transcripts, including prompts, responses, tool activity, and metadata such as verified user information and timestamps, through the existing Compliance API interface. Organizations can retrieve audit and monitoring data for these products without standing up separate logging infrastructure. Source

3. Miro Runs Content Review and Project Tracking on Claude Cowork

Anthropic. A new Anthropic customer story details how collaboration software company Miro uses Claude Cowork to automate content review, project tracking, and strategic planning. Miro reported that 99% of influencer content went live within 48 hours across three 2026 events, up from 64% previously, and that weekly project tracking dropped from more than 90 minutes to roughly 10 minutes. Staff also use Cowork as a thought partner to test half-formed problems before drafting deliverables. Source