1. The fanfic world built an AI detector to catch cheaters. It can't tell them from everyone else Fanfiction writers on AO3 spent the past week running each other's stories through a tool built to detect text generated by Claude.

2. Alibaba Tags Claude Code as High-Risk Software, Reportedly Citing Backdoor Fears Alibaba has classified Claude Code as high-risk software and barred employees from using it, according to a report from TechCrunch. The reported reason is backdoor risk.

3. Bhavin Turakhia Put $30 Million of His Own Money Into an AI Office Suite. It's His Fifth Enterprise Bet. Bhavin Turakhia is funding his newest company himself, to the tune of $30 million.

In Brief

• NVIDIA opens its AI factories to outside capital partners NVIDIA launched a program inviting financial partners to fund large-scale, multi-tenant GPU clusters as demand shifts from training to continuous inference. The company frames these "AI factories" around keeping accelerators highly utilized to make token-scale services profitable.

• Amazon, Adobe, and Citi cap employee AI usage over runaway costs Leaks from Amazon, Adobe, Atlassian, and Citi show employers throttling internal AI tool access because per-seat inference bills are climbing faster than expected. Workers report hitting usage limits mid-task.

• Google's electricity use rose 37% in 2025 on data center expansion Google reported a 37% jump in electricity consumption for 2025, driven by AI data center construction. The company is buying clean energy to offset emissions but the gap between power draw and carbon-free supply widened.

• Meta released Pocket, a text-to-minigame app Meta quietly shipped Pocket, an experimental app that generates and shares interactive mini games from text prompts. The product extends Meta's push into prompt-driven, AI-generated consumer content.

• Privacy advocates ask FTC to keep monitoring Musk's X Advocacy groups urged the FTC to reject Elon Musk's request to end privacy oversight of X, arguing the platform's AI data practices pose serious risk to Americans. The petition targets Musk's bid to lift a consent-decree monitor.

• Campaign launches to protect the legal right to run local AI A new advocacy site, Right to Intelligence, argues users should retain the ability to run AI models on their own hardware without regulatory or vendor restrictions. The effort surfaced on Hacker News amid debates over model access controls.

• Researchers release AgenticSTS to test long-horizon agent memory A new paper introduces AgenticSTS, a bounded-memory testbed that assembles each agent decision from typed retrieval instead of appending all prior context. The design isolates the effect of individual memory components on long task chains.

• A developer automated dating outreach with OpenClaw and Claude Code Ben Guez built a script using OpenClaw, Claude Code, and Instagram trial accounts to auto-message potential matches, filling his DMs with responses. The setup shows off-the-shelf agent tooling applied to personal outreach at scale.

• TechCrunch surveys Chrome and Safari alternatives as browser competition shifts to AI TechCrunch compiled current browsers challenging Chrome and Safari, noting the fight now centers on built-in AI features rather than search defaults. The roundup covers privacy-focused and AI-native options.

Read the full edition →