Good morning. Anthropic’s reported IPO numbers lead today’s edition, followed by a new privacy feature in ChatGPT, a dispute over how a documentary portrays an AI agent breach, and a debate about how readable neural networks really are. Our explainers cover video, music, robotaxis, smart glasses and vendor lock-in.
Daily DigestDigest
Daily Digest: Anthropic's draft IPO numbers, ChatGPT Locked Chats, a documentary dispute and a neuron debate
AI stories that mattered on October 22, 2026: Anthropic's reported IPO filing, ChatGPT Locked Chats, a documentary row and a debate on reading neural networks.
Today in 12 stories · October 22, 2026
- 01Markets & Economy
Anthropic's draft IPO prospectus shows a $42 billion loss and $518 billion in compute commitments
Reuters reports 2025 revenue near $4.6 billion; most of the loss is a non-cash charge. The document is not public.
- 02Product Launches
ChatGPT adds Locked Chats, a PIN- or biometric-protected section with its own memory
A screen-privacy feature, not encryption; resetting the PIN deletes locked content.
- 03AI Safety & Alignment
Hugging Face's CEO says Netflix's 'AI Gone Rogue' distorts the OpenAI agent breach
Clem Delangue says the film omits the missing containment and monitoring behind the incident.
- 04AI News
Goodfire researcher claims neural networks are easier to read than assumed; peers call it a hunch
A 26,000-word post argues neurons are cleaner than believed; critics want systematic evidence.
- 05Big Tech
OpenAI models head to Google Cloud: what the deal means for big tech rivalry
OpenAI says its models and Codex workflows will be offered through Google Cloud's Model Garden.
- 06Interpretability
Interpretability explained: neurons, features, circuits and superposition
The core concepts and tools behind reading a neural network.
- 07Chart of the Week
Chart of the Week: where Anthropic's reported $518 billion in compute commitments goes
A text chart of the reported split across six partners.
- 08AI PCs, Phones & Devices
AI smart glasses explained: how they work, what Snap Specs offers and what to weigh
Types of smart glasses, Snap's reported Specs details and a privacy checklist.
- 09Video Generation
How AI video generators work: from diffusion to temporal consistency
Latent diffusion, transformers over space and time, and a quality checklist.
- 10Audio & Music
How AI music generation works: audio tokens, diffusion and stems
Audio codecs, two model families and licensing checks.
- 11Autonomous Vehicles & Drones
Why robotaxis are hard to scale: the economics behind driverless ride-hailing
Vehicle cost, remote support, utilization and how to read expansion claims.
- 12Business & Enterprise
A practical guide to avoiding AI vendor lock-in with a multi-model, multi-cloud setup
Where lock-in hides and a six-step checklist.
Sources
This story was reported and written in our own words from the sources below. Primary sources are preferred.
- Anthropic lost $42bn in 2025 as revenue grew 12-fold, Reuters reports — thenextweb.com
- ChatGPT now lets you lock your chats — techcrunch.com