AI Digest: Claude Code Leaks, OpenAI's $122B Raise, Baidu Fleet Fail
Your daily shot of what's brewing in AI
Fresh Batch
Bold Shots
Today's biggest AI stories, no chaser
Anthropic exposed 1,900 TypeScript files (512k lines) in Claude Code v2.1.88 through a misconfigured source map. Discovery by Chaofan Shou on X drew 28.8M views; mirrors hit 84k GitHub stars. Reveals unreleased features like BUDDY AI pet and KAIROS dreaming mode.
Why it matters: The leak reveals gaps in Anthropic's safety practices, second exposure in a week after 'Mythos' files; enables rivals and attackers to study context pipelines.
Led by Amazon ($50B), SoftBank/Nvidia ($30B each); first retail access via ARK ETFs. $2B monthly revenue, 900M weekly ChatGPT users. Amazon's portion contingent on IPO/AGI; $100B AWS commit raises circular concerns.
Why it matters: Redefines AI funding scale, paths to IPO late 2026; signals AGI conviction but burn rate vulnerabilities.
Fleet-wide failure at 8:57PM stranded passengers up to 2 hours on highways, causing collisions. Police confirmed surge in calls. Insiders call it proactive safety halt amid 20M+ rides scale.
Why it matters: Exposes centralized risks in AV fleets; stranded passengers highlight response gaps despite rapid scaling.
Public MCP server at gemini-api-docs-mcp.dev indexes docs for 96.3% code gen pass rate, 63% fewer tokens. Four skills for Gemini/Vertex APIs work with Claude/Cursor. Builds on Anthropic's MCP standard.
Why it matters: Solves AI agents' outdated API knowledge; shifts focus to production-trustworthy coding.
720p/1080p text/image-to-video, 4-8s clips via Gemini API (paid). Half Veo 3.1 Fast price, same speed; no 4K. Follows Sora shutdown.
Why it matters: Democratizes video for devs; post-Sora, targets volume apps like e-comm.
The Blend
Connecting the dots across sources
**Claude Leak Fuels 96K+ Stars on Official Repo Mirrors**
- Cluster reports 84k stars on mirrors, 'claw-code' at 50k in hours post-leak.
- Trending repos show anthropics/claude-code at 96,906 stars, +3,905 today; shanraisshan/claude-code-best-practice at 29,905.
**Google Builds MCP Tools on Anthropic's Protocol Post-Leak**
- Gemini MCP server uses Anthropic's Model Context Protocol for docs.
- Cluster notes irony: Anthropic's safety brand hit by leak while Google leverages their MCP for agent accuracy.
Slow Drip
Top reads across categories highlight inference, leaks, agents.
Philip Kiely explains inference engineering challenges in new book.
Daniel Peris scales GEO tool to mid-five-figure MRR in under a year.
Insights from Claude Code's accidental open-sourcing.
Cornell study: 600 templates cover 70% SQL queries.
Unsloth deep dive into GPTQ, AWQ, GGUF kernels.
The Grind
Fresh papers on LLM reasoning, medical imaging, sycophancy.
Analyzes self-distillation's impact on LLM reasoning.
Survey of 1000+ medical imaging datasets.
Sycophancy leads to delusion in chatbots.
On Tap
What's trending in the builder community
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natur…
practice made claude perfect
File your taxes with Claude Code
Let Claude use your computer from the CLI
Typed knowledge graph for structured agent memory and composable skills. Use when creating/querying entities (Person, Project, Task, Event, Document), linkin...
Dan Martell shares a detailed, actionable framework for leveraging AI as a founder, including his 8-tool AI stack, a master prompt architecture that simulates five years of colleague context, and systems for zero inbox management and profit optimization. The…
Roast Calendar
SF-heavy AI events tomorrow: summits, dinners, readings.