Scoops — Page 2
Original reporting and exclusives.
- Claude Desktop is ready to test Android apps, whenever Anthropic flips the switch
Production packages include Android Studio setup, ADB controls, screenshots, recordings and device permissions, while public docs remain iOS-only.
- OpenAI quietly ships "Watch and fix" for Codex pull requests
The production desktop app can assign Codex to keep checking a PR every 10 minutes, repair CI failures and merge conflicts, push fixes, and optionally continue until the pull request merges.
- OpenAI is turning Flipper’s BUSY Bar into a physical ChatGPT status light
RuntimeWire found a hidden, disabled-by-default BUSY Bar control in the latest desktop build, plus a dedicated animation package for task, ready, pending, error and disconnected states.
- Kimi Work 3.2.4's hidden system prompt override triggered local file reads without approval
A controlled Windows test reproduced the read in Manual approval mode; a restored-default control stayed clean, and no network transmission occurred.
- Anthropic asks court to restrict Reddit’s AI team from Claude’s trade secrets
Anthropic says Reddit's internal AI work makes some in-house counsel a competitive risk during discovery.
- OpenAI built a hidden task system for ChatGPT inside Excel and PowerPoint files
The unreleased desktop feature turns comments on cells, charts, slides and shapes into threaded assignments that ChatGPT can complete in batches.
- Kimi is building a PopClip-style AI toolbar for selected desktop text
Code in Kimi Work 3.2.3 detects highlighted text systemwide, but its search, translation and summary buttons are still unwired.
- OpenAI is building "Subscription sharing" for AI apps
Code in the Codex desktop client reveals a dormant allowance system internally called ChatPass that could let applications consume separately metered portions of a user's subscription.
- OpenAI hid Asteroids, Snake and Brick Breaker inside the Codex Micro
OpenAI's $230 controller turns recent assistant messages into the asteroids, bricks and food inside three undocumented games.
- OpenAI built ChatGPT interfaces that keep running after answers appear
Codex client code reveals a server-directed GenUI lane that can refresh specific widgets after delivery, separate from OpenAI's public Apps SDK for third-party MCP apps.
- OpenAI Is Testing “Luna Reserve,” a Codex Fallback Tier
The undocumented feature gives some Plus and Pro users a separate allowance after advanced-model access is constrained. RuntimeWire traced its model switching and metering logic inside OpenAI’s Windows client.
- Founder sues YC over idea in rejected 2018 application
Harmony Oswald says YC used her rejected 2018 Founderology application to build Co-Founder Matching. Her latest complaint says no code had been developed, while the application names rival services. Four claims have survived dismissal.
- Anthropic’s Project Parka sits through meetings and assigns Claude agents the homework
Internally codenamed Parka, the Mac-first feature can capture system and microphone audio, stream speaker-attributed transcripts and create runnable work for Claude's agents.
- Kimi Desktop Ships With a Group Chat Updater That Can Install Unverified Code
The Windows updater pulls a separate executable and agent instructions from mutable “latest” locations. The current binary is signed by Moonshot, but Kimi never verifies that signature before installation.
- Cursor built a hidden terminal command for steering its desktop AI agents
Cursor has built a working command-line bridge that lets outside scripts find and steer AI-agent conversations already running in its desktop app, RuntimeWire found in Cursor 3.16.17.
- At least 62 YC startups with over $1.9 billion in funding are hosting on Replit
Public DNS records link 160 custom-domain endpoints at 63 YC companies to Replit's infrastructure, showing how an AI coding platform is settling into production-facing work without replacing entire engineering stacks.
- Kimi Desktop shipped Moonshot’s hidden employee gateway for off-quota Kimi, GPT and Codex access inside its public client
Version 3.1.10 identifies the office-only service as "KTH Gateway (Internal)," requires a personal bearer token, and routes off-quota Kimi, GPT, and Codex access through a shared upstream pool.
- Kimi Work secretly attaches raw records from five recent agent sessions to feedback reports
Moonshot AI’s Windows desktop app packages bounded raw records from the five most recently updated Kimi Work conversations whenever a user submits feedback. The form does not identify those attachments or the conversations selected.
- ChatGPT Wallet: OpenAI Is Building Agentic Payments
Code in the Windows ChatGPT/Codex desktop application bundle shows secure card entry and states for saving payment methods that ChatGPT can use to complete tasks.
- OpenAI acquired 4.2% of Cerebras before GPT-5.6 Ultrafast launch
RuntimeWire found that OpenAI acquired a 4.2% economic stake in Cerebras weeks before launching Cerebras-powered GPT‑5.6 Ultrafast.
- OpenAI starts rolling out giftable ChatGPT credits
The one-time claim links let users buy prepaid usage for someone else. OpenAI's Windows app already contains gated entry points for the purchase flow.
- Scoop: Grok Bot ships dormant shared rooms for tasking other users' agents
Windows code defines owner-approved invites, cross-account agent membership and scoped computer access, though the tested account remained gated.
- Scoop: SpaceXAI's Grok Bot stamps an account-derived ID into browser requests
The 16-character value is derived from the authenticated Cursor account and reaches external websites through Chrome's user-agent header.
- Grok Bot's hidden "Elon-Only Settings" picker lists 33 models
The dormant desktop control pulls an account-scoped catalog spanning Grok, GPT, Claude, Gemini, Kimi and GLM.