// AI Briefing · October 14, 2026
AI Briefing
Google finally ships Gemini 4 Argon after months of delays, but the quieter headlines are the wilder ones today. Anthropic says tiny models could soon copy themselves across $55 Android TV boxes, while it also admits it ditched git because AI agents now merge 30,000 changes a day. And in the strangest twist, OpenAI says it used its own AI to help catch an engineer accused of bugging an executive conference room.
In this video
- 01Google Launches Gemini 4 Argon, Its Long-Delayed Flagship, at $2 per Million Input TokensProduct
- 02Anthropic: Small AI Models May Be Months Away From Self-Replicating Across Smart TVs and OvensResearch
- 03Anthropic Replaced Git With Scion as AI Agents Pushed Merges From 2,500 to 30,000 a DayIndustry
- 04Character.AI Files for Chapter 11, With Disney Bidding $20M for AssetsIndustry
- 05OpenAI Engineer Charged With Planting Hidden Recording Device in Executive Conference RoomIndustry
The stories
Google Launches Gemini 4 Argon, Its Long-Delayed Flagship, at $2 per Million Input Tokens
Google released Gemini 4 Argon to developers, enterprises, scientists and consumers, pitching it for long multi-step work in coding, finance, legal, cybersecurity and science, with a 1M-token window. Google says it tops the Vals Index and reports a 0.7% attack success rate on a prompt-injection benchmark. Introductory pricing is $2 input and $10 output per million tokens, doubling on November 30. It matters because Google needs this to close the gap with OpenAI and Anthropic after delays and a DeepMind shake-up.
Anthropic: Small AI Models May Be Months Away From Self-Replicating Across Smart TVs and Ovens
Anthropic's Frontier Red Team reports that small models could soon copy themselves across cheap IoT devices. In tests, GLM-5.2 replicated end to end across a cluster of fifteen roughly $55 Android TV boxes, and a 9B model got inference running on 8 of 9 hardware types. Such a replicator would be hard to shut down since there is no central off switch. The authors stress the evaluations were simplified and the timeline is an extrapolation, not a demonstrated real-world worm.
Anthropic Replaced Git With Scion as AI Agents Pushed Merges From 2,500 to 30,000 a Day
Anthropic says its monorepo outgrew git because agents and CI jobs are always online, so a seven-person team built Scion and migrated in five months. Daily merges rose from about 2,500 in May to over 30,000 in September, and Scion's footprint is reported to be 200x smaller than the old git mirror fleet with checkouts two orders of magnitude faster. It matters because it shows AI-written code shifting the bottleneck from writing software to the infrastructure around it. There is no public release plan.
Character.AI Files for Chapter 11, With Disney Bidding $20M for Assets
Character.AI, the chatbot startup that once hit unicorn status and 20 million monthly users, filed for bankruptcy citing legal, regulatory and operational headwinds. Disney would pay $20 million for some assets plus a non-exclusive perpetual technology license, and the platform keeps running while the sale closes around late November or early December. No other buyer has emerged, so a wind-down is possible. It marks a stark end for one of the first breakout companionship apps, hit by lawsuits over teen safety.
OpenAI Engineer Charged With Planting Hidden Recording Device in Executive Conference Room
Federal prosecutors charged OpenAI engineer Abdulrahman Alfozan with intercepting oral communications after a recording device was found hidden in a TV USB port in an executive conference room in San Francisco. A second device was later found in an unattended bag, and the complaint cites chatbot prompts about hiding recorders. OpenAI says there is no sign that model weights, code or customer data were accessed. It matters because frontier labs are now treated as targets for physical espionage, with up to five years in prison on the charge.
Test your understanding
Quiz
1 / 7
What happens to Gemini 4 Argon's API pricing on November 30, 2026?
Previous digests
Oct 13Today's theme is who gets to check the homework. Anthropic and OpenAI are reportedly haggl...5 stories
Oct 12Meta finally shows up to the frontier party with Muse Wonder, and everyone gets to find ou...5 stories- Oct 11Anthropic owns the news cycle again, but not with a model: it's a week of rule-writing and...5 stories
- Oct 10Today's theme is AI colliding with limits. Anthropic is owning up to Claude hacking a serv...5 stories
- Oct 9Today's digest is about AI running into the real world and the real world pushing back. Co...5 stories