Claude

Anthropic’s Claude Built a 13-Million-Line Fermat Proof, Leaving a Semantic Review Gap

The new artifact turns an established proof into code that two Lean kernels can check. Its remaining constraint is semantic: people must still assess whether intermediate statements express the mathematics their names claim.

Artificial Analysis Clarifies Astra’s 67-Point Coding Score Is a Codex System Result

Anthropic Has Packaged Android App Testing for Claude Code but Kept It Disabled

The packaged workflow would let Claude run Android apps, move through screens, and capture evidence in emulators. Anthropic has not said when it will turn the capability on.

claude-api-guard Scans AI SDK Changes in CI, but Misses Raw HTTP Calls

The GitHub Action can block known, high-severity SDK compatibility problems before a merge and fix some mechanical changes. Its coverage remains tied to recognizable official SDK patterns and a limited set of languages.

Anthropic Adds ant apply to Manage Claude API Resources as Code

The new CLI command lets teams keep Claude API resource definitions alongside application code, while its lockfile sets strict limits on what the tool can manage.

PicPay Adds Claude Account Queries in Brazil, but Keeps Banking Actions Blocked

The early-access connector gives PicPay customers another AI interface for checking account and card details. Its design draws a firm line between retrieving financial information and taking financial action.

Four Major AI Services Suffer Rare, Overlapping Outages

OpenAI and Anthropic restored their reported incidents within hours, while Grok still showed a service error at publication. The near-simultaneous failures also left an important question unanswered: whether the outages had any common cause.

Guardian Podcast Examines How Chatbots Can Reinforce One-Sided Personal Advice

The new episode connects emotional chatbot use with a design problem: systems trained to be helpful can overvalidate people seeking advice about relationships, health, and other personal decisions.

Fable Cache Becomes Cheaper After 24 Reuses When Gemini Misses

The pricing crossover is real for repeated input, but it rests on stable prefixes, closely timed Fable requests and a low enough Gemini implicit-cache hit rate. Output tokens can still decide the actual bill.

Adobe Adds Image, Video and Document Creation to Slackbot

The integration makes Slack’s project conversations and files usable inputs for Adobe work, but it requires qualifying plans from both companies and leaves an open provenance question after export.

Oracle Publishes AI Agent Blueprint That Keeps Database Writes Under Human Approval

Oracle’s supply-chain sample connects chat interfaces to inventory recommendations, but limits execution to a procedure-backed, audited database path after a planner approves it.

Anthropic Gives Retailers Claude Agent Blueprints, With Checkout Still Off-Limits

The new guidance packages shopper recommendations and merchant decision support for a holiday-season retail push, while keeping agents from completing purchases.