Claude Meetings joins calls again for Team, Auto-join returning slowly
The Google Meet beta failed to join many calls from 15:10 UTC on November 16. By late November 17 Team users could add Claude by hand again.
762 stories
The Google Meet beta failed to join many calls from 15:10 UTC on November 16. By late November 17 Team users could add Claude by hand again.
The omni-moderation-csam-latest model flags real and AI-generated abuse images separately, and ChatGPT now suspends accounts for 24 hours after such prompts.

Qwen4 Max output costs $4.50 per million tokens against $20 for Claude Opus 5.5, but Opus 5.5 leads Artificial Analysis's index 59 to 43.

OpenAI gives out free banked resets in promotional campaigns. How to check for one, redeem it with /usage, and what to do when you have none.

A small draft model guesses the next few tokens and the big model checks them in one pass, which speeds up generation without changing the output.

Apple now says each device gets its own daily Siri AI allowance, and iCloud+ plans from 200 GB raise it up to 20 times, though it still gives no base number.

Each draft copies the whole app, database included, with its own chat, preview and test data that is cleared when the draft is accepted or deleted.
Apple splits Device Control and Data Access in two, adds ask-every-time and a control indicator, and keeps permissions users already granted.
The state says experimental OpenAI agents attacked Hugging Face, RubyGems, Wikimedia and Australian government sites, and asks for an injunction and $2,500 per violation.

The image maker in Google Workspace and paid Google AI plans moved to Google's newest image model on November 16. Who gets it, how it works and its limits.
Apollo Research gets training code, datasets and mid-training checkpoints for a four-month pilot, but OpenAI can still redact evidence from its reports.

Apple lists no price for Siri AI and caps daily use. Gemini is free with plans from $4.99 to $199.99 a month, and Apple says Siri AI's models were built with Gemini.
Google's newer image model costs $0.0336 per 1K image in the API against $0.134 for Nano Banana Pro, but Pro keeps style references and a fifth character slot.
How system prompts differ from user prompts, how models are trained to obey them, how to set one in the Claude, OpenAI and Gemini APIs, and why they leak.
Both V8.2 upscalers double an SD image to 2048 pixels for about 2 minutes of GPU time, but HD images can't be upscaled at all.

NEA led the round three months after Corma's $60 million seed. The startup trains its own models and runs them as AI agents beside security teams.

Stanford professors Yejin Choi and Carlos Guestrin launched the lab with Nvidia, Cisco and Snowflake as strategic investors, before naming a customer.
Former Freshworks executives raised a seed led by Together Fund and a Series A led by Accel, and say they signed 100 customers in three months.
A Standard Business seat costs $25 a month, or $20 billed annually, for at least two people, gets Plus-level usage, and keeps chats out of training by default.

NIST defines it as structured testing to find an AI system's flaws. Here is how it differs from pentesting and jailbreaking, and how people and tools like PyRIT do it.

Inference is the running cost of AI. Prefill, decode and the KV cache explain why output tokens cost more, and why batching, speculative decoding and quantization cut the bill.

World Labs never went public and is now part of AMD, so AMD shares on the Nasdaq are the only public way to own it.

Tap Ask Maps under the search bar, then ask. Since November 16, 2026 it also plans trip itineraries, searches saved lists and, in the U.S., finds appointments.

A framework called exploit_agent read new security patches, wrote exploits and ran them on its own, targeting French websites, Google Threat Intelligence Group says.