Weekly AI Digest: Anthropic’s $1.5B Copyright Bill and Google’s going bananas

Week 36, 2025

This week’s big story is Anthropic’s settlement to pay $3,000 per pirated book used in training. It’s the first legal outcome in the piracy cases, but it’s not the same as authors being paid when AI is trained on copyrighted work legally. The legal side is messy, full of fine details, and it will be worth watching how the AI law space develops. Expect more lawsuits against companies training large models.

Also this week: OpenAI keeps buying up smaller players, and Anthropic closed its Series F at a massive $183 billion valuation.

Also, Google’s new image generation model (nano banana) is absolutely mindblowing.

Let’s go for the digest.

Anthropic pays $1.5B to authors

Anthropic agreed to a $1.5 billion settlement with authors over pirated book datasets used to train Claude. Each work gets about $3,000, and the company must delete the dataset. Anthropic denies wrongdoing but avoids trial risk while Apple was separately sued in California over similar claims.

While this looks like a major precedent for AI copyright, the case is about piracy — books used without permission. The broader question of training on copyrighted works remains unresolved.

Read more on Wired

Anthropic raises $13B, at a post-money valuation of $183B

..

They are probably going to pay the aforementioned lawsuit (if they ever pay) with a chunk of the huge capital amount they’ve just raised.

In a world where the race for becoming the top AI company is heating up every week, Anthropic just raised a large sum of money for hiring top talent, research and product developement.

Read more on the Anthropic Blog

OpenAI acquires Statsig, reshuffles leadership

OpenAI is buying product-testing startup Statsig for $1.1 billion. Founder Vijaye Raji becomes CTO of Applications, with other leadership shuffles including Kevin Weil, currnt CPO becoming VP of OpenAI for Science.

OpenAI is doubling down on controlled experimentation and product scaling, with acquisitions showing how fast the company wants to industrialize its app ecosystem.

Read more on TechCrunch

🖼️ Google’s Gemini launches “Nano Banana”

Google rolled out “Nano Banana,” a new image-editing model inside Gemini. It handles likeness retention, multi-image fusion, and conversational editing. Reportedly attracted 10 million users and processed 200 million images in weeks.

Image editing is becoming a wedge use case for mass adoption. The results of the tool are very impressive. Check this picture I snapped of a street in my city:

I’ve asked Gemini (with the new update) to turn this image intro bright daylight:

Absolutely crazy! Only by looking careful can w spot some errors in the image, but it’s impressive.

Read more: Google Blog and try it on Gemini App

GPT-5 vs GPT-4o blind test

In terms of technical performanca, we know that GPT-5 largely beats GPT-4 (94.6% vs 71% on math, 74.9% vs 30.8% on coding, 80% fewer hallucinations) but is this reflected on user experience? That’s what this website wants to put to the test.

Users noticed improvements but also voiced fatigue with “new model” hype, voicing concerns about the experience within the new GPT-5 ecosystem.

Benchmarks look strong, but public trust and perception are now as important as performance as GPT-5 rollout showed us.

Read more on Venture Beat.

OpenAI’s collective alignment update

OpenAI shared results from its “Collective Alignment” work, gathering ~1,000 global perspectives on model specifications. Feedback largely aligned, but disagreements pushed updates in safety and transparency rules.

Experiments and research on governance and risk will be so important as time runs. How models behave should increasingly be influenced by public input.

Read more on the OpenAI blog.

The week things moved in all dimensions of the AI field. On the legal battles defining ethics of training, on big tech racing to capture distribution through products, and the public shaping AI rules of the road. The hype curve may be flattening, but the truth is that there are so many quesrions yet to answer that the interest shouldn’t die for many years to come.

As always, stay curious,

– Ivo