Anthropic released Claude Fable 5.2 and Mythos 5.2 while Nvidia's board opened a probe into CEO Jensen Huang, and Meta, Google, ByteDance and others added smaller launches and policy news.
98 stories ranked · updated Nov 9, 22:34 UTC
01
Anthropic releases Claude Fable 5.2 and Mythos 5.2, with Mythos limited to verified organizations
Anthropic launched Claude Fable 5.2 for demanding Reasoning model: A model that thinks step by step before answering. Slower, but much better at maths, code and tricky questions. and long-horizon Agent: An AI that doesn't just answer but takes steps on its own to finish a task: searching the web, running code or using apps. work, and Claude Mythos 5.2 with the same capabilities for verified organizations only. Both are priced at $8 per million input and $40 per million output Tokens: The small chunks of text a model reads and writes, about three-quarters of a word each. Prices and limits are counted in tokens., down 20% from 5.1.
Nvidia board opens probe into Jensen Huang over affair with suspected Chinese spy
Huang admitted an affair with a woman US agencies classify as a suspected Chinese informant, and Nvidia's board launched an independent investigation. Nvidia shares fell about 4-5% and Huang pulled out of at least one event.
Meta releases Muse Ember, a low-cost small model for agent tasks
Meta's Muse Ember targets repetitive Agent: An AI that doesn't just answer but takes steps on its own to finish a task: searching the web, running code or using apps. work such as tool calls, routing and data extraction. Meta says it is cheaper per Tokens: The small chunks of text a model reads and writes, about three-quarters of a word each. Prices and limits are counted in tokens. than Haiku and GPT-6 Luna, at about 10 cents per million input tokens.
Fireworks AI says hackers accessed customer datasets, model weights and billing data
AI Inference: Running a trained model to get an answer. Training builds the model once; inference happens every time someone uses it. provider Fireworks disclosed that attackers accessed some customers' datasets, incomplete model weights and billing information over the past month. Two customers told Forbes they were affected.
Google Gemini reportedly sends personal Drive files and emails to human annotators
A GitHub repository and a Hacker News thread allege that Gemini routes personal Google Drive files and emails to human annotators. The claims are unverified in the items provided.
OpenAI details Control Room, a system to track and stop model workloads
OpenAI published Control Room, which tracks model workloads across its research infrastructure and can quickly stop them when safety or security concerns arise.
Trump says DOJ threatened a top AI executive over bad activities
In a 60 Minutes interview, Trump said the Justice Department threatened a top executive at one of the big AI companies over bad activities and that the person 'totally changed'. He also said the risk of AI is 'managed'.
Google expands TPU 8 family with two specialized variants
Google added the TPU 8tx, which keeps data encrypted during processing, and the TPU 8i Max, which places two chips on one package to double compute and memory. The move extends its push against Nvidia in AI hardware.
ByteDance Seed introduced Seedream 5.5, which it says keeps images stable across up to 100 editing rounds without color drift. It also generates native RGBA images with transparent backgrounds.
UnconfirmedNot confirmed yet: A hint that a model is coming before it's announced: a name spotted in a lab's systems, an anonymous test model, or code that adds support for it. Usually right, sometimes not.NotableNotable release: A meaningful new model: a new size or version, or a strong free-to-download model from a known lab.
NotableNotable release: A meaningful new model: a new size or version, or a strong free-to-download model from a known lab.OpenOpen model: Its files are public (open weights), so anyone can download it and run it on their own computer, usually for free.
Meituan, the Chinese services company, released LongCat-2.5, a free-to-download AI assistant that can read images and handle long coding and computer tasks.
NVIDIA
Meta
Google
ByteDance Seed