1. Mark Zuckerberg Is Training a Digital Copy of Himself to Talk to His Own Employees Meta is building an AI avatar of its CEO, the Financial Times reported.

2. Berkeley Gamed Every Top AI Agent Benchmark. Stanford Found the Public Never Trusted the Scores. A research team at UC Berkeley set out to exploit the most widely cited AI agent benchmarks. They succeeded on every one.

3. Three AI Companies Hit Commercial Milestones This Week. None of Them Train Models. Three companies at different layers of the AI stack, none of them model builders, reported commercial milestones in the same week. Vercel signaled IPO readiness.

In Brief

• OpenAI CRO Sends Internal Memo on Locking In Users Against Anthropic Chief revenue officer Denise Dresser circulated a four-page strategy memo to OpenAI staff Sunday. The document pushes the company to build a moat around its AI products and accelerate enterprise growth, citing how easily customers switch between providers.

• Sam Altman Targeted in Second Attack at San Francisco Home Two suspects were arrested Sunday after a shooting at Altman's Russian Hill residence. Police charged both with negligent discharge after surveillance footage showed a vehicle near the property.

• Trump Officials Reportedly Push Banks to Test Anthropic's Mythos Model Administration officials are encouraging financial institutions to pilot Anthropic's restricted Mythos model, per TechCrunch. The push directly contradicts the Department of Defense's recent designation of Anthropic as a supply-chain risk.

• Microsoft Tests OpenClaw-Style Autonomous Bots for 365 Copilot Microsoft is experimenting with OpenClaw-like features inside its Copilot assistant, The Information reports. The company wants 365 Copilot to run autonomously around the clock, completing tasks on behalf of users. Corporate VP Omar Shahine confirmed the tests.

• 70+ Civil Rights Groups Warn Meta Facial Recognition Glasses Will Endanger Vulnerable People The ACLU, EPIC, Fight for the Future, and more than 70 other organizations oppose adding facial recognition to Meta's smart glasses. They argue the feature would put abuse victims, immigrants, and LGBTQ+ individuals at direct risk.

• Apple Tests Four Designs for Upcoming Smart Glasses Apple is prototyping four distinct form factors for a smart glasses product line, per TechCrunch. The project scales back earlier plans that called for a full range of mixed and augmented reality devices.

• Hacker Breaches a16z-Backed AI Phone Farm, Tries to Post Anti-a16z Memes A hacker gained backend access to Doublespeed, a startup that uses phone farms to flood social media with AI-generated influencer accounts. The attacker attempted to post memes calling backer a16z the "Antichrist."

• Anthropic Dominates Conversation at HumanX Conference Attendees at San Francisco's HumanX AI conference made Claude the central topic of discussion, per TechCrunch. Anthropic drew more attention than any other company at the event.

• College Instructors Call LLM Use the Most Demoralizing Problem in Teaching An Ars Technica essay frames widespread ChatGPT use among students as the hardest challenge college instructors now face. Detection remains unreliable, and academic integrity frameworks have not kept pace.

• LG AI Research Releases EXAONE 4.5 Open-Weight Vision Language Model LG AI Research published EXAONE 4.5, integrating a visual encoder into its EXAONE 4.0 framework for native multimodal pretraining. Training emphasized document-centric data aligned with LG's enterprise application domains.

Read the full edition →