Good morning. It’s an Anthropic-heavy day: the company shipped Sonnet 5, launched a new Claude Science app, got its Mythos and Fable export ban lifted — and, less flatteringly, was caught hiding steganographic markers in Claude Code prompts. Meanwhile Google has a faster Nano Banana and Etched crossed $1B in chip orders. Coffee first.
Claude Code is secretly marking your prompts. A developer reverse-engineering Claude Code v2.1.196 found that the binary subtly alters the apostrophe in “Today’s date” and the date separators based on the user’s timezone and API endpoint hostname — a steganographic fingerprint apparently designed to detect model distillation via Chinese AI lab infrastructure. The HN thread is unhappy less about the intent than the concealment: one commenter noted Anthropic could have shipped an explicit telemetry field with documentation and release notes, and another pointed out this is the second such mechanism found in Claude Code this year. A separate thread compared it to Google’s “don’t be evil” arc, compressed into eighteen months.
Sonnet 5 lands, but the pricing math is odd. Anthropic released Claude Sonnet 5 at $3/$15 per million tokens (intro pricing $2/$10 through August 31), positioning it as agentic capability approaching Opus 4.8 for less. TechCrunch frames it as undercutting GPT-5.5 and Gemini 3.1 Pro, but the HN reaction keeps returning to Anthropic’s own cost-per-task chart: above medium effort, Opus 4.8 is