Claude Writes Crash Proofs for 123 Libraries
Claude's machine-checked proofs found hundreds of bugs fuzzers missed, but they only cover RISC-V binaries and can take hundreds of millions of tokens.
Read the storyProduct · General assistants
An AI assistant for analysis, writing, coding and working with documents.
Get AI Secret’s free daily briefing, including Claude and the wider AI industry.

An AI assistant for analysis, writing, coding and working with documents.
Official product information checked .
Briefs, articles and relevant sections, newest first. Explore older coverage below.
More coverage
Claude's machine-checked proofs found hundreds of bugs fuzzers missed, but they only cover RISC-V binaries and can take hundreds of millions of tokens.
Read the storyPlus: Humans Were the Bug, Giants Wants Free Chatters to Pay
Anthropic's new API version turns temperature, top_p and top_k into 400 errors on every model, while the 2023 version stays alive for holdouts.
Claude Duet briefs users daily, works in its own cloud browser and asks before paying, but checks connected accounts only every few hours.
Plus: Anthropic Pulled Up the Ladder, OpenAI Copied Meta's Ad Machine
Opus 5.5 hands flagged security requests to Opus 4.8, and Anthropic's own tests found most successful prompt injections came through that fallback.
DeepSeek's V4.1 Flash tops open models on Code Arena WebDev at a tenth of Kimi K3's cost, though Claude Opus 5 still leads most benchmarks.
Replit's CEO said it dropped Anthropic. Two days later, Claude was still available, and OpenAI had simply become the default most users keep.
Replit's CEO says it dropped Anthropic for OpenAI, but people close to the deal deny it, and Replit's own blog and docs still run Claude.
Plus: Epic Delayed Star Wars for Claude, China's Humanoids Are Just Toys