AI Briefing

Today the agent era stopped being a demo and started acquiring paperwork. OpenAI handed developers the exact harness that runs Codex, Visa and Mastercard began building an ID check for AI shoppers, and California became the first state to say labs can't be the only ones grading their own homework. Meanwhile DeepSeek is quietly heading for the public markets, and IBM and NASA dropped an open model that reads the surface of the Moon.

01ProductOpenAI ↗

OpenAI Just Gave Every Developer the Same Harness That Runs Codex

OpenAI put its Agents API into public beta, exposing the managed harness behind Codex as a service any developer can call. One API call creates a production agent with a model, instructions, tools, MCP servers and an optional sandbox environment, and OpenAI handles the hard parts: durable sessions that survive for days, context compaction, tool orchestration and subagent fan-out. There is no extra API fee beyond model tokens and paid tools. The interesting shift is that agent infrastructure, the thing every startup has been rebuilding from scratch for two years, is now a commodity layer.

02IndustryPYMNTS ↗

Visa and Mastercard Are Building a Background Check for AI Shoppers

Ant International, Visa and Mastercard announced they are aligning on a Know-Your-Agent interoperability framework so an AI agent verified on one network is recognised across all of them. The three had each shipped competing protocols already, Visa's Trusted Agent Protocol, Mastercard Verifiable Intent and Ant's Agentic Mobile Protocol, and this folds them into shared rules covering cross-network operator traceability, common certification requirements and continuous transaction monitoring. The work runs through BuildFin.ai, a platform convened by the Monetary Authority of Singapore. The number driving all of it: the group projects agents will orchestrate $3 to $5 trillion of consumer commerce by 2030.

03PolicyQuartz ↗

California Says AI Labs Can No Longer Be the Only Ones Grading Their Own Homework

Governor Gavin Newsom signed two bills creating the first framework in the United States requiring independent third-party audits of AI systems. AB 1405, from Assemblymember Rebecca Bauer-Kahan, establishes a state registry for AI auditors with standards for independence, transparency and integrity, and bars anyone without a state registration number from conducting a covered audit starting January 1, 2029. SB 813, from Senator Jerry McNerney, sets up a framework for independent verification organisations to assess models for compliance with state law. The long runway matters less than the precedent: safety evaluation is becoming a licensed profession rather than a blog post from the lab that built the model.

04FundingQuartz ↗

DeepSeek Is Heading for the Public Markets at Roughly $75 Billion

DeepSeek has engaged CITIC Securities to prepare an initial public offering on Shanghai's STAR Market and aims to begin the listing process this year, according to Reuters. Deal size, timing and target valuation are all undetermined, but a pre-IPO financing round reportedly values the Hangzhou company around 500 billion yuan, about $75 billion, up from the roughly $50 billion post-money valuation it carried after raising $7.4 billion in June. Chinese firms must retain a securities company for pre-IPO tutoring before filing, so hiring CITIC is the first formal step. For the lab that made cheap frontier training a geopolitical story, this is the pivot from private capital to public markets.

05Open SourceAIwire ↗

IBM and NASA Open-Sourced a Foundation Model That Reads the Surface of the Moon

IBM and NASA released the NASA-IBM Lunar Foundation Model, one of the first publicly available foundation models built for scientific exploration of the Moon. It was trained on a large curated lunar observation dataset drawn from decades of multi-instrument sensor data, petabytes of it, that scientists previously had to sift through by hand or process with low-resolution, task-specific tools. The stated goal is supporting a sustained human presence on the Moon by turning that archive into something queryable. It is also a useful reminder that the foundation-model pattern is escaping language entirely and landing in domains where the bottleneck was never text.