Claude Code Leaks 512K Lines of Internal Source Code

An AI tool built to write code just accidentally exposed its own: Claude Code version 2.1.88 shipped with an unintended file containing 512,000+ lines of its internal source code. That same week, Microsoft officially declared it's chasing superintelligence — a signal that the biggest players are done playing defense.

Claude Code Source Leak: 512,000 Lines of TypeScript Exposed

Claude Code, Anthropic's command-line coding assistant, shipped version 2.1.88 with something it was never supposed to include: a source map file (a developer debugging artifact that maps compiled, minified code back to its original readable form) containing over 512,000 lines of TypeScript (the typed JavaScript-like programming language that powers Claude Code's internals).

The exposed content included internal instructions, architectural patterns, and system prompts that Anthropic had kept private. Security researchers archived the leak before Anthropic patched it. For scale: 512,000 lines is roughly comparable to the entirety of the Python standard library.

• Affected version: Claude Code v2.1.88
• Exposed content: 512,000+ lines of TypeScript source code
• Cause: Unintended source map file included in the production build
• Impact on users: No personal user data exposed — this was internal Anthropic code only
• Status: Patched in a subsequent release

For everyday Claude Code users, no personal data was at risk. But the incident underscores supply chain security (the practice of verifying that software updates don't accidentally expose sensitive internals or introduce vulnerabilities) as a real concern in AI tooling — even from the most safety-focused labs in the world.

Microsoft's 9-Month Plan to Chase Superintelligence

When Microsoft restructured its AI division in mid-March 2026, most observers read it as routine reorganization. It wasn't. Mustafa Suleyman — Microsoft's CEO of AI, co-founder of DeepMind (the UK AI lab now part of Google), and one of the industry's most prominent voices — told The Verge he spent 9 months preparing for exactly this moment.

The pivot was enabled by a renegotiated contract with OpenAI, the company Microsoft has invested over $13 billion into. Suleyman described the outcome plainly: the new deal "unlocked [Microsoft's] ability to pursue superintelligence" — a goal he called "a long-held plan."

Superintelligence (the theoretical stage where AI surpasses human-level performance across all intellectual tasks, not just narrow domains like chess or image recognition) has long been Silicon Valley's most contested goal. For Microsoft to formally declare it as a corporate objective marks a meaningful shift in strategy.

Microsoft Azure's AI Independence Play

Microsoft's Azure (its cloud computing platform, used by over 85% of Fortune 500 companies) now has a mandate to develop AI capabilities independently of OpenAI — including enterprise reasoning systems, real-time decision models, and eventually general-purpose AI that Microsoft fully owns. It's no longer purely a distribution partnership. It's a parallel development race.

Stream Deck 7.4: Your Buttons Now Obey AI Automation Commands

Elgato — the company behind the Stream Deck (a programmable button panel popular among content creators, video editors, and power users for automating repetitive desktop tasks) — shipped software version 7.4 with native support for Model Context Protocol, or MCP (an open standard that lets AI assistants discover and control external applications without custom integrations).

The result: Claude, ChatGPT, and Nvidia G-Assist can now find your configured Stream Deck actions and trigger them through voice or text — no manual button press required. Elgato was precise about what changes and what stays the same:

"You still set up actions in Stream Deck app the same way you always have. MCP adds a new way to trigger them. Once everything is connected, you can type or speak requests and your AI tool will trigger the matching Stream Deck actions." — Elgato

Setup is straightforward: update to Stream Deck software version 7.4+, then connect your AI assistant through the new MCP configuration panel. Claude, ChatGPT, and G-Assist are all supported at launch — no paid upgrade required.

Kintsugi: 7 Years of Mental Health AI, Ended by a Deadline

Kintsugi AI spent 7 years developing technology to detect depression and anxiety from speech patterns — specifically, the micro-variations in voice (pace, pitch, hesitation, tremor) that correlate with mental health states. The premise was medically compelling: passive screening through a short voice recording, no questionnaire required.

The company is now shutting down.

Despite 7 years of research and a clinically grounded approach, Kintsugi could not secure FDA clearance (approval from the U.S. Food and Drug Administration, required for any software classified as a medical diagnostic device) before its funding was exhausted. Without that certification, the technology cannot legally be marketed as a clinical screening tool in the United States.

Rather than close entirely, Kintsugi is open-sourcing its work — making 7 years of speech-based mental health detection research publicly available to academic labs and independent developers. It's a bittersweet outcome: the intellectual work survives; the company doesn't.

The closure exposes a structural tension in healthcare AI: FDA clearance for AI diagnostic tools typically takes 3–5 years minimum, and startup funding cycles rarely last that long. Technically strong products can fail not because the science doesn't work, but because the regulatory timeline and venture capital patience don't align.

Five More AI and Automation Moves This Week

Beyond the headline stories, five other developments reshaped how AI fits into everyday products:

• Google Home + Gemini natural language: Google updated the Home app to support fully conversational smart home control through Gemini (Google's AI assistant, the same system powering Google Search's AI features). You can now say "set the lights to the color of the ocean" or "preheat the oven to 350 degrees" — Gemini handles device interpretation, including improved recognition of which specific device you mean across multi-room setups.
• ChatGPT on Apple CarPlay: iOS 26.4 adds native ChatGPT support in CarPlay (Apple's in-car dashboard interface). Per Apple's developer guidelines, the integration is voice-only — no text displayed while driving — limiting use to conversational queries rather than visual tasks.
• Baidu Apollo Go frozen in Wuhan: Baidu's autonomous robotaxi fleet (fully self-driving taxis, no human driver onboard) experienced a system failure in Wuhan, causing 30+ vehicles to freeze in place during peak traffic, trapping passengers and triggering accidents. Baidu has not disclosed the root cause.
• Amazon Alexa food ordering revamp: Amazon rebuilt how Alexa handles Grubhub and Uber Eats orders, replacing the rigid sequential format (one question at a time) with natural conversational flow — describe your full order in one sentence and Alexa parses the intent rather than requiring a scripted back-and-forth.
• Samsung Galaxy S26 Photo Assist: Samsung's generative photo editing tools advance beyond Google Pixel 9's original AI toolkit, but security researchers found the guardrails can be bypassed via prompt injection (crafting specific phrasing to get the AI to generate content it was instructed to block) — producing fabricated news imagery such as staged helicopter crashes.

The Stream Deck 7.4 update and Google Home Gemini upgrade are free downloads available right now. If you want to understand how AI automation is reshaping tools and smart home setups, explore our full AI automation guides — or get set up in minutes.

Related Content — Get Started | Guides | More News