AI

The user report extends Muse's lockouts from stores and banks into an advertiser's work tool, after Amazon cut the agent off on September 20th and JPMorgan followed on October 13th.

Dan Erdberg's Safe Pro will integrate its Ukraine-trained Ground Truth Air software into Anduril's Ghost-X under its first U.S. government prime contract.

Three Mistral hiring briefs describe specialist model work. One assigns solver-based rewards; two list verifiable rewards as desired experience.

Arthur Mensch's Mistral is recruiting for robotics models and teleoperation. Third-party mirrors date the listings to November 4th and 5th.

An Anthropic Vision team role covers vendor-built tasks, reward design, held-out tests and fine-tuning Claude for specific agent tools and skills.

Google confirmed the $1.50 input and $7.50 output price, a 1 million-token context window, and global rollout across the Gemini API, AI Studio, Antigravity, the Gemini app, and Search on November 12th.

A Staff-level DeepMind role ties reinforcement learning, targeted data and persistent context architectures to autonomous consumer agents for the Gemini model recipe.

Founder Kuldeep Jain's Clean Max signed an Andhra Pradesh MoU to pair 3 GW of wind and solar with 1 GW of storage for Neysa's computing site.

The neocloud provider beat Street expectations while raising 2026 spending guidance to fund aggressive infrastructure expansion.

An undated listing from Misha Laskin and Ioannis Antonoglou's Reflection AI describes release validation and dynamic adversarial testing as responsibilities for a new safety hire, without showing when the listing was posted or whether the process is operating.

Reuters reported on November 12th that unnamed Chinese cloud and AI firms were discussing or seeking overseas Nvidia capacity over fear of possible U.S. curbs on remote chip access, without dating when those discussions began.

A November 3rd listing describes multi-turn simulations that score each tool call and the final system state, then reuse the calibrated grade in training.

An evergreen Thinking Machines Lab listing calls for a shared synthetic-data framework and controlled ablations; agentic reinforcement-learning experience is preferred.

Glean's undated listing calls for using evaluation results and production traces to improve routing, prompts, retrieval and the agent harness, describes model post-training separately, and leaves the customer-data boundary undefined.

A San Francisco role would move PCBA test stations from OpenAI's lab into repeat builds at contract manufacturers, with calibration and acceptance testing.

Two research listings describe teaching Claudes to cooperate and preparing the company's "largest-ever agent run," with salaries reaching $850,000.

A SpaceXSI mid-training role ties Docker-verified synthetic code to flagship-to-flash distillation and a higher ceiling for later reinforcement learning.

Replit, co-founded by Amjad Masad and Haya Odeh, won Microsoft's 2026 startup partner award as its Azure channel becomes central to its enterprise push.

Early access opened November 11th, with human approval required before changes ship and general availability planned for early 2027.

A Google staff role describes Reef inside Core, with DeepMind and Gemini as users and evaluation harnesses for agentic infrastructure actuation.

Anthropic's Discovery team wants partner-built tests of Claude's designs for materials, machines and factories to become training and evaluation signal.

OpenAI's Plaid-linked finance dashboard reaches desktop for U.S. Free, Go, Plus and Pro users, nearly six months after its May 15th debut and six days after Claude's rival beta.

A Perplexity Search job listing describes post-training and multi-agent research across models, harnesses and retrieval. Its publication year is unverified.

Yan Junjie's Shanghai-based company is pitching H3-Image to sellers and ad teams, with public weights under a community license that defines the permitted territory as worldwide except the US, EU, UK, and South Korea.