Digest · August 15, 2026
Claude Week — August 15, 2026
Welcome to the first issue. Every week we collect what actually changed in the Claude world — status, releases, limits and the ecosystem around it — and link you onward to the sources.
Following all this is one thing. If you would rather be good at it, we teach that at ClaudeMaster.
The biggest things worth knowing
- Claude now watermarks the text it writes — Models released from August 2 embed an invisible mark in generated text, worldwide. · TechCrunch
- Claude Code auto mode becomes the default — Pro, Max and Team accounts switched over on August 14. · TechCrunch
- Claude for Teachers launches — A K-12 version built around state academic standards. · Anthropic
- Sonnet 5 promotional pricing ends August 31 — Standard pricing takes effect September 1. · ClaudeLog
- Two major incidents in three days — August 13 and 14 both hit multiple components — full timelines in our archive. · ClaudeNetwork
Stories
- 1
The watermark, and the week of pushback that followed
Anthropic updated a support page confirming that models launched from August 2 embed an invisible statistical mark in generated text, with C2PA provenance metadata on files. The driver is Article 50 of the EU AI Act, enforceable from the same date, but the policy applies worldwide with no opt-out. The objection that gained the most traction is narrower than the headlines: text a person wrote themselves can pick up the mark if Claude is only asked to proofread it. Anthropic notes the reverse also holds — a missing watermark does not mean text was human-written.
Forbes
- 2
Auto mode is now the default, and what that actually changes
From August 14, Claude Code stops asking for approval at every step for Pro, Max and Team accounts. It still stops for actions judged irreversible, destructive or reaching outside your environment. Auto mode was introduced as a test in March; the change is the default flipping, not new capability.
TechCrunch
- 3
A detection API is coming, but the terms are not set
An Anthropic engineer confirmed a text detection API that customers will be able to call themselves, and a follow-up FAQ described the mechanism as a version of DeepMind's SynthID-Text. Pricing, rate limits and access tier are still unpublished, which leaves the interesting question open: open detection invites iteration until the mark disappears, while gated detection makes independent verification harder.
explainx.ai
Trending tools
- claude-meter — Tracks how fast you burn through your plan's usage window. · claude-meter
- ClaudeLog — A dated running log of Claude changes and announcements. · ClaudeLog
- Releasebot — Aggregates official vendor release notes, including Claude Code. · Releasebot
- PulseMCP — One of the larger MCP server directories, with weekly ecosystem notes. · PulseMCP
- Glama — The biggest MCP index by server count. · Glama
- Claude Status — The official component status feed everything here is built on. · Anthropic
What we're reading
- Every Claude usage limit change, dated · explainx.ai
- Where Claude's model lineup stands in August 2026 · AIToolsReview
- Users push back on watermarking · TechCrunch
- What the watermark does and does not prove · Gizmodo
- Where to find MCP servers in 2026 · AutomationSwitch