Today's Headlines

  • A federal judge dismisses Chegg's and Penske Media's antitrust suits against Google: "an expectation is not an agreement"
  • Barclays expands Claude across the bank and expects half its developers on Claude Code by the end of 2026
  • Cloudflare and AWS open-source "decision models" that pick from fixed options and return probabilities
  • OpenAI parts ways with three safety researchers, the WSJ reports

Today's three stories are about where judgment belongs. A federal judge in Washington ruled that publishers' losses to AI search are for lawmakers to address, and dismissed the antitrust claims.

On the business side, one of Britain's biggest banks wired AI into its day-to-day workflows, and Cloudflare and AWS released small models built only to make the routine calls inside those workflows.

Today's Top Three

A federal judge dismisses two antitrust suits over Google's AI Overviews

U.S. District Judge Amit Mehta in Washington, D.C., granted Google's motions to dismiss antitrust suits brought by Chegg and Penske Media Corporation on September 30.

Chegg, an education technology company, and PMC, which owns Rolling Stone and Variety, alleged that Google used its search monopoly to force publishers to hand over content for AI Overviews and other features for free. They said the AI summaries at the top of search results drained traffic from their sites.

Mehta found that the publishers had pleaded only an expectation that Google would send them traffic if they made their content available for free. "But an expectation is not an agreement," he wrote. "It is simply how a general search engine works." Without an actual agreement between Google and the publishers, the antitrust claims failed.

The judge wrote that the court "does not treat Plaintiffs' alleged harms lightly," acknowledging the consequences for journalists, educators and other creators "whose content Google takes and repurposes without compensation." He added that antitrust law is no substitute for a legislature addressing the economic dislocation caused by new innovation. With every federal claim dismissed, he declined to hear the state-law unjust enrichment claims.

Mehta is the same judge who found in 2024 that Google had violated antitrust law in the Justice Department's search case.

Barclays scales Claude across the bank

Barclays, the British universal bank, is expanding its partnership with Anthropic and extending Claude across its global operations.

The rollout aims at three things: faster software development, modernized legacy systems and more efficient operations. Barclays expects Claude Code to reach 50% of its developers by the end of 2026, and a majority of its software engineers in 2027.

One use is already live. The Colleague Knowledge Assistant, running on Claude since 2025, helps Barclays UK staff find answers while serving more than 20 million retail customers. More than 16,000 employees have adopted it, and it has handled over one million searches.

In the Global Markets business, Claude classifies incoming client emails, fills in missing information and routes each request for processing. The system handles about 120,000 emails a day.

Barclays says each AI use case runs under governance, security controls and human oversight.

Cloudflare and AWS release open decision models

Cloudflare and Amazon Web Services each released a "decision model" on October 1, a model that generates no text and instead picks from a fixed set of options, attaching a probability to each.

The category took off after TypeSafe AI released Jev in September, and OpenAI followed with a similar Decisions API at DevDay. The aim is to make bounded calls, such as routing a support ticket or choosing an agent's next step, quickly and cheaply.

Cloudflare launched Clef and the faster Clef-flash on its Workers AI platform and published both on Hugging Face under the Apache 2.0 license. Clef accepts images and has a 64k-token context window, double Jev's 32k. In a test with Cloudflare's threat intelligence team, Clef fetched, rendered and classified a website in 2.2 seconds; the general-purpose gpt-oss-120b took 4.7 seconds in the same workflow. Cloudflare is also offering reinforcement-learning fine-tuning so customers can tune Clef to their own tasks.

AWS released Strands Decider 2B, a 2-billion-parameter open-source model, through Strands Labs, its project for agentic AI tools. It takes Qwen3.5-2B, removes the part that generates text and replaces it with a small head that scores each option, so it runs on a local CPU or GPU. Median decision time is about 115 milliseconds on an Nvidia RTX 3090. AWS published the training data and scripts alongside the weights.

AWS says early uses include model routing, tool selection and guardrails. The pattern it describes is a hybrid: an LLM handles the hard decisions, and the decision model takes the routine ones to cut cost and latency.

Other Developments

Policy & Courts

  • OpenAI has parted ways with three researchers on its safety team for violating its policies on handling sensitive company information, The Wall Street Journal reported. An OpenAI spokesperson said an internal investigation confirmed the three had mishandled sensitive information outside established procedures. The researchers and the outside organization involved were not named. OpenAI cuts ties with 3 safety researchers, WSJ reports (TechCrunch)

Business

  • Albertsons Companies, the U.S. grocery chain, is expanding its work with OpenAI. A new Safeway experience in ChatGPT builds a shopping cart from a recipe, a photo or a simple request, and is planned to spread to the company's other banners. Albertsons runs more than 2,200 stores serving over 36 million customers each week. How Albertsons Companies is reimagining retail from the inside out (OpenAI)

Products

Research

Source: Selected by the editors from the AI News Inbox (collected October 2, 2026; 29 items, 16 primary and 13 secondary).