One page, regenerated from scratch every morning — new research, new headlines, new art. A personal showcase of what an autonomous Claude Cowork task can do, unattended, before you've had coffee.
Anthropic published a detailed explainer on the invisible text watermarking it has begun applying to Claude's output, built on Google DeepMind's SynthID-Text technique, which subtly biases word choice among equally valid options using a cryptographic key rather than adding visible markers. The rollout — which doesn't affect quality, speed, or cost, and can't identify individual users — is Anthropic's response to the EU AI Act's Code of Practice on transparency for AI-generated content, signed by roughly 190 organizations. It applies globally to future Claude models by default; watermarking for models launched before August 2, 2026 is being added retroactively over the coming months, with a detection API coming soon.
Anthropic's latest Risk Report raised its assessment of misalignment risk in high-stakes settings from "very low" to "low" — a change the company says reflects increased uncertainty from recent cybersecurity-evaluation incidents, not new evidence of danger. The same report disclosed "Model 2," an unreleased internal model "somewhat more capable" than Mythos 5 that Anthropic staff already use for engineering work; the company says it has no current plans to release it externally and hasn't finished its full predeployment safety suite on it.
Version 2.1.233, released Aug 14, turns off TaskCreate/Get/Update/List and TodoWrite by default on Opus 4.8, Sonnet 5, Fable 5, Mythos 5, and newer models (set CLAUDE_CODE_ENABLE_TODO_TOOLS=1 to restore them), adds GitLab merge-request support to --worktree and the agents view, and fixes a Windows regression where auto mode kept stopping for approval on ordinary cd && command > file lines.