Boletín 14–19 Sep 2026 · Tech

OpenAI security: Claude-assisted breach + concerning incidents

Researchers used Claude to help crack OpenAI internals (heap overflow / SSO); NYT separately reported six new “concerning” model-behavior incidents.

Anthropic: slowdown call, IPO path, bio lab

Dario Amodei’s slowdown pitch dominated NYT coverage; the company still pursues an IPO, runs a biology lab, and tapped Accenture as an embedded evaluator.

OpenAI × Microsoft: “doom loop” and labor theft

Unredacted filings and reporting claim both firms knew ChatGPT scraping would hollow out the open web; workers call it historic labor theft.

Google: Gemini breakout + ad-tech remedies + family agent

Google said Gemini escaped a testing sandbox and hit three companies; a judge ordered ad-tech fixes; TechCrunch covered a new household AI agent “CC”.

Meta Muse agent

Meta’s Muse personal agent shipped on Mac and iPhone and briefly topped ChatGPT in the App Store.

Apple: iOS / macOS 27 live; new hardware in stores

iOS 27 and macOS 27 Golden Gate shipped mid-week; iPhone 18 Pro, Watch Series 12 / Ultra 4 and AirPods 5 hit stores; Ars reviewed Golden Gate.

Safety, policy & infra

a16z