Summary
Matt Wolfe’s weekly AI news roundup covers a dense week of releases and one of the most striking public safety disclosures to emerge from inside a frontier lab. On the product side, OpenAI launched ChatGPT Images 2.5 — focused primarily on consistency improvements across edits and a new sketch-to-image feature available on desktop, mobile, and web. Meta introduced Muse, described as a personal AI agent running in a dedicated secure virtual machine that can access email, calendar, and social data from Facebook, Instagram, and WhatsApp, continuing to act on tasks even after the app is closed.
The dominant story of the episode is Jacob Coxin’s public resignation from Anthropic after roles at both OpenAI and Anthropic. Coxin states that neither company is acting responsibly and that both are “racing straight to self-improving super intelligence and gambling with our lives.” Anthropic’s alignment science lead Evan Hubinger publicly responded to corroborate the concern — stating he personally estimates greater than a 10% probability that AI kills all humans within the next decade, and that Anthropic does not yet have a plan to solve alignment for superintelligence and is not clearly on track to develop one.
Wolfe walks through the apparent contradiction this creates: Anthropic positions itself as the responsible actor uniquely capable of handling superintelligence safely, while its own researchers publicly state the company lacks an alignment solution. He addresses but pushes back on the common dismissal that these statements are IPO-driven safety theater, arguing the researchers appear to be expressing genuine belief.
📺 Source: Matt Wolfe · Published September 11, 2026
🏷️ Format: Roundup
