01
PHP / Unite. AI ↗
PHP has released security updates for two vulnerabilities that can let an unauthenticated attacker execute code on a web server, with installations serving web requests affected. Researchers at Calif and Wiz used frontier AI models in the discovery work, while PHP maintainers coordinated the fixes. The patched releases are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35; the PHP project says updating is the only complete solution.
02
Anthropic ↗
Anthropic has introduced Claude Duet, an agent that checks connected apps, prepares daily briefs and keeps working in its own virtual machine when the user steps away. Built around Claude Opus 5.5, it asks for approval before actions such as sending messages, making payments or deleting information. The research preview starts with US Pro and Max subscribers through a waitlist, shares their existing usage limits and is available on iOS, Android and the web.
03
OpenAI / Axios ↗
OpenAI and Google Cloud have announced that OpenAI frontier models and Codex workflows will be offered through Google Cloud’s Model Garden and Marketplace. Axios reports that customers will be able to use existing cloud spending commitments, with general availability planned in the coming weeks. The agreement extends OpenAI’s distribution across the three major US cloud platforms and lets enterprise customers access its models through infrastructure they already use.
04
NVIDIA ↗
Nvidia has introduced DSX POD, a reference design for modular Vera Rubin AI factories assembled and tested before shipment. The company says installation can fall from six months to as little as 14 days at sites where power, cooling, networking and supporting infrastructure are already prepared. Partners including Dell, HPE, Lenovo, Schneider Electric and Vertiv are working toward worldwide deliveries in early 2027, shifting much of the integration work from the construction site to the factory.
05
Recursive ↗
Recursive has released Flint, an Apache 2.0 coding model with 40 billion total parameters and 2.5 billion active per token. The company says its Eureka Machine research system helped develop the model’s architecture, training and inference stack, enabling a small team to build it in under four months. Recursive reports more than 125 tokens per second on an M4 Pro MacBook Pro with its optimized engine, offering developers a local option that keeps code on their own hardware.
06
Unite. AI / Meta ↗
Meta reported third-quarter revenue of $64.6 billion, up 26%, and approved a new publicly listed class of non-voting shares to support capital raising, employee compensation and potential acquisitions. It narrowed its 2026 capital expenditure outlook to $135 billion–$145 billion and said it would keep investing aggressively in AI. The proposed Class C structure preserves existing voting power, giving Meta another way to finance its expansion while maintaining Mark Zuckerberg’s control.
07
Walmart ↗
Walmart is introducing dedicated Agent Profiles so shoppers can see an AI agent’s purchases separately, require approval or set a spending limit, and revoke access. The profiles are due to launch in the next month, while a developer beta of the Walmart Agent Gateway is being built to let agents shop through a direct connection rather than navigating pages designed for people. Walmart says agents will see the same prices as human shoppers and will not receive their payment-card details.
08
IBM / Hugging Face ↗
IBM has released Granite Speech 5.0 alongside updates to its Turbo CTC and speaker-attribution models, all under Apache 2.0 licenses. The new two-billion-parameter model supports transcription in six languages, expanded translation and keyword lists of up to 2,000 terms for names and specialist vocabulary. The smaller 470-million-parameter Turbo CTC model adds multilingual transcription and formatted output, while the updated Plus model improves speaker labels and timestamps for meetings and calls.
09
Bloomberg via Techmeme ↗
AI research startup Flapping Airplanes has raised $1 billion at a $6 billion valuation, Bloomberg reports. The company is pursuing methods to train AI models with less data and compute, alongside work to make chips more efficient. The funding backs a research approach focused on reducing the resources needed to build AI, rather than relying solely on larger training runs.
10
Glean / Unite. AI ↗
Glean has made its headless agent builder generally available, letting developers create, test and publish agents from Claude Code, Cursor and Codex. Agent configuration and instructions live in files that teams can version and review, while published agents run on Glean’s managed platform with its enterprise permissions, knowledge graph and audit trail. The feature supports Auto mode agents, giving engineering teams a code-based workflow without requiring them to build a separate enterprise runtime.