Anthropic is putting its models and engineers behind the people who defend power grids, water systems and transport networks. The new Cyber Mission starts with a Critical Infrastructure Defense Program whose 11 founding partners include CrowdStrike, Palo Alto Networks, Accenture and Rockwell Automation, and with OSS Scanner, a free service that runs regular security scans from its strongest models for enrolled open-source projects and sends maintainers a proof of concept and a suggested fix for what it finds.

Anthropic's yearly policy rewrite pulls in two directions at once. It drops the blanket ban on personalized vote and campaign targeting, narrowing the elections rules to deceiving voters and disrupting democratic processes, while consolidating bans on fake-account influence campaigns into one deceptive-activity section. The weapons rules now explicitly cover guidance and control software and arming drones, surveillance limits are spelled out in more detail, and there is a new ban on sustained, needless abusive behavior toward the models themselves. The updated policy takes effect on November 12.

Meta and Sierra, joined by Genesys, Instinct, Rocket, Shopify, Stripe and Walmart, proposed an open standard for how personal AI agents identify themselves to businesses and what they are allowed to do once inside. Sessions run on OAuth and carry across channels, customers choose read-only or write access, and each company decides whether an agent works through its website, its APIs or an agent of its own. Nothing has shipped yet: the v0.1 specification is promised later in October, with a reference implementation to follow.

Google opened its SynthID Detector to everyone, worldwide in English, after a year in which the portal was limited to journalists and media professionals. Anyone can now upload an image, video or audio file and check it for the invisible watermark Google embeds in AI-generated media. The detector also reads marks from partners OpenAI, NVIDIA and Kakao, with Apple named as coming next. Google says SynthID has now been applied to more than 180 billion images and videos, plus 240,000 years of audio.

Anthropic is expanding its Cyber Verification Program, which vets security researchers for access to Claude's cyber capabilities.

At a joint event in San Francisco, Jensen Huang and Satya Nadella put NVIDIA's RTX Spark superchip - a Grace CPU plus a Blackwell GPU with up to 128GB of unified memory - into Windows laptops built to run agents locally. Laptop preorders are open ahead of October 16 availability, compact desktops follow in November, and Microsoft made its Execution Containers for agents generally available.

Atlassian and OpenAI are expanding their partnership to bring AI deeper into Jira, Confluence and team collaboration.

Anthropic's smallest model takes a big jump: Haiku 5.5 starts at $0.10 per million input tokens for prompts up to 100k, roughly 75% cheaper to run than Haiku 4.5, and it is the first Haiku with an adjustable effort setting. Anthropic also halved Sonnet 5.5 cache-read prices and is adding a monthly API credit for Max and Team subscribers.

OpenAI is bringing GPT-6 to the Chat tab for the 1.2 billion people who use ChatGPT each week, with a new Intelligent UI that answers with interactive visuals, charts and tools built on the spot instead of plain text. Paid tiers run on GPT-6 Sol from today; Free and Go tiers follow tomorrow on GPT-6 Luna.