Courts Side With Google on AI Traffic Theft

Two of the more prominent names in publishing — education platform Chegg and Penske Media Corporation, the parent company of Rolling Stone — took Google to court last year claiming its AI Overview search feature was strangling their web traffic, according to The Verge.

The core argument: Google was using publishers' content to generate AI-written summaries at the top of search results, which meant users got their answers without ever clicking through to the original source. That diverted traffic, which meant fewer readers, which meant less advertising revenue.

US District Judge Amit Mehta dismissed both cases on Wednesday. His reasoning was direct: publishers have an expectation that Google will send traffic their way if they make content available, but an expectation is not a contract. That is simply how a search engine has always worked.

Mehta added that the court was not unsympathetic to publishers' situation, but that antitrust law is not the right tool to address what is essentially a policy question about the economics of new technology. That, he wrote, is a job for lawmakers.

The ruling does not mean Google is off the hook with Congress. But for now, the legal route has closed. Publishers navigating traffic losses from AI search have no immediate recourse through the courts.

What to watch: whether the ruling accelerates direct licensing deals, like the pilot programme Google reportedly runs with around 100 publishers for AI Overviews contributions.

Google has announced Gemini 4 Argon, its latest frontier model — meaning a model operating at the upper edge of what current AI systems can do — targeted at complex tasks in software engineering, legal work, finance, and cybersecurity, according to The Verge.

The catch: it is not available to the public yet. Access is currently limited to a "set of trusted cyber defenders," and Google says it is going through the US government's voluntary pre-release review process before a broader rollout. Google is also reportedly already using the model in its own internal workflows.

Google shared a benchmark chart showing Gemini 4 Argon outperforming models from OpenAI and Anthropic. Those benchmarks were run by Google. No independent evaluation has been published.

The announcement came one day after OpenAI's developer conference, which itself featured new model and agent releases. The timing is not subtle.

Honest read: Worth watching, not worth acting on. The model is not publicly available, the benchmarks are self-reported, and the access restrictions are real. Check back when independent testers have had a go.

What it is: A training framework for building large AI models, specifically designed for a type of model architecture called Mixture-of-Experts (MoE), where only a small portion of the model's components are activated for any given input rather than running everything at once.

What's new: Previous versions of OLMo-Core used a system called FSDP (Fully Sharded Data Parallelism), which repeatedly gathered and resharded model weights — the numerical values that encode what a model has learned — across GPUs during training. OLMo-Core 3 switches to a different approach that keeps expert components resident on their assigned GPUs and routes data to them instead, avoiding that repeated shuffling.

How it works: Experts are spread across GPUs using expert parallelism. Model layers are split across GPU groups using pipeline parallelism. The optimizer state — the data used to calculate and apply weight updates during training — is distributed rather than duplicated. For each input token, only four experts are activated, regardless of how large the full expert pool grows.

What it can't do: In a preliminary test on eight NVIDIA B300 GPUs, a 47-billion-parameter version processed 52,000 tokens per second per GPU, compared with 19,400 using the earlier implementation — about 2.7 times the throughput. But this is preliminary. Trillion-parameter scale benchmarks are early-stage, and real-world training runs involve coordination costs that benchmarks on eight GPUs may not fully capture.

Who should care: Researchers and smaller labs priced out of frontier model training — this is specifically built to make that more accessible.

Try it / read it: OLMo-Core 3 on Hugging Face

The Open TTS Leaderboard, available now on Hugging Face, ranks text-to-speech (TTS) models — software that converts written text into spoken audio — using objective, automated metrics rather than human voting panels.

It tracks three things: how accurately the audio reproduces the original words (measured by word error rate), how fast the model generates audio, and how closely a cloned voice matches the original speaker's voice. Over 8,000 TTS models are available on Hugging Face; this leaderboard focuses specifically on open-weights models, which are underrepresented on existing ranking sites.

If you are choosing a TTS model for a project, or just curious which open-weights options are worth testing, filter by language and check the Pareto plots — they show which models balance accuracy and speed without needing to read every row.

Access it here: Open TTS Leaderboard

Worth knowing: The leaderboard measures intelligibility and speed objectively, but does not measure naturalness or how pleasant a voice sounds — those still require human ears to judge, and scores here won't tell you which model sounds best.

OpenAI has published details of a coordinated campaign it disrupted in July, in which operators systematically extracted protected reasoning — the internal chain of logic a model works through before producing an answer — from its models without authorisation, according to OpenAI's own disclosure.

The technique, called adversarial distillation, involved manipulating model interactions rather than breaching any database. OpenAI attributes a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi. OpenAI has shared its findings with the Frontier Model Forum so other AI developers can strengthen their own defences.