Here are today's news for 7th of October: OpenAI will roll out invisible text watermarks to eligible ChatGPT and Codex outputs in the EU over the coming weeks, responding to the AI Act’s machine, readable transparency rules; Anthropic expanded its Claude Startups program, offering approved companies founded within five years or funded in the past two a free year of Claude Team for up to five Premium seats, plus a one, time $1,000 Claude API credit; Meta and Sierra, with Walmart, Stripe and other partners, announced the open Personal Agent Protocol on Oct; Google launched EmbeddingGemma 2 on October 6, an open, weight, 740, million, parameter model built on Gemma 4 for on, device search; and Google announced Gemini 4 Argon on September 30, beginning a limited rollout to trusted cyber defenders through its Fairwind Program. The first and most important one: OpenAI will roll out invisible text watermarks to eligible ChatGPT and Codex outputs in the EU over the coming weeks, responding to the AI Act’s machine, readable transparency rules. Let's dive into the details. From OpenAI: OpenAI will roll out invisible text watermarks to eligible ChatGPT and Codex outputs in the EU over the coming weeks, responding to the AI Act’s machine, readable transparency rules. Its textGrain system subtly steers word choices to create a detectable statistical pattern; API users worldwide can opt in for select models, but it remains off by default. The approach could help establish AI, content provenance, though OpenAI says editing weakens detection: replacing 25% of words with synonyms cut detection to 17% in tests. From Anthropic: Anthropic expanded its Claude Startups program, offering approved companies founded within five years or funded in the past two a free year of Claude Team for up to five Premium seats, plus a one, time $1,000 Claude API credit. The credit expires after six months; Team is free only for startups new to the plan. The move lowers early, stage teams’ costs to build with Claude as AI providers compete to win startup developers. From Meta, Walmart, Stripe, and Sierra: Meta and Sierra, with Walmart, Stripe and other partners, announced the open Personal Agent Protocol on Oct. 6, aiming to standardize how AI agents identify themselves and act for customers on business sites. The effort could make agent, driven shopping safer and easier to manage: customers grant read, only or write access, while businesses set limits. It uses OAuth and can work through websites, APIs such as MCP or OpenAPI, or a company’s own agent; a v0.1 specification is planned for later this month. From Google: Google launched EmbeddingGemma 2 on October 6, an open, weight, 740, million, parameter model built on Gemma 4 for on, device search. It maps text, code, images, video and audio into a shared 768, dimensional space, enabling searches across media without sending data to the cloud. Released under Apache 2.0, it can run in modular configurations; Google says the full version uses about 567MB of active RAM on a Pixel 11 Pro. From Google: Google announced Gemini 4 Argon on September 30, beginning a limited rollout to trusted cyber defenders through its Fairwind Program. The model is designed for complex software, legal and finance work, and can autonomously find, validate and patch critical software vulnerabilities. Google says defenders will get access without cyber guardrails, while it strengthens safeguards before broader release, a move that could speed vulnerability fixes but raises the stakes for controlling powerful AI capabilities. That's all for today's AI news briefing. Have a great day!