Meta just poached 8 OpenAI researchers with $100 million offers while federal judges greenlighted training LLMs on copyrighted books and Google dropped a free tool that could ctrl+alt+delete Anthropic's coding advantage overnight.

Sheesh.

We're talking about the wildest talent war Silicon Valley has seen in years.

Plus legal victories worth billions and a Senate deal trading $500 million federal funding for a 5-year state regulation freeze.

Oh, and Salesforce's CEO claims AI now does half their company's work. Which we're calling complete BS on.

This week in AI developments had everything. 

Let's get into this week's AI feast, shorties. 👇

Two federal judges in California just ruled that Anthropic and Meta can legally train their AI models on copyrighted books. These are the first major US court decisions addressing how copyright law applies to AI training.

Judge William Alsup determined that Anthropic's use of millions of copyrighted books to train Claude was totally legal for books the company actually purchased. They literally bought books, ripped off covers, and scanned them.

But it’s not a closed-book case. (Lolz) 

Anthropic still faces claims over a pirated library containing over 7 million books. So they're not completely clear yet.

In a separate case, Judge Vince Chabara sided with Meta against 12 authors including Sarah Silverman who sued over Llama AI training. The judge basically said the authors did a terrible job making their case. 

Yikes. 

What it means: 

AI companies just got their first major legal shield, though this will be climbing the court system. 

This creates a clear playbook: buy content legally, prove you're not reproducing verbatim copies in your model, and you're (apparently) golden.

Expect a massive surge in AI companies purchasing training data instead of scraping it. Like….. They got the cash. 

The Supreme Court will ultimately decide this in the U.S.

OpenAI got slapped with a lawsuit from startup IO over their new hardware device name in partnership with former Apple designer Jony Ive. 

This turned into a full social media spectacle.

Sam Altman went kinda nuclear and posted private email receipts on Twitter showing IO's Jason Rugolo had pitched OpenAI for a $10 million investment.

 Altman declined, saying OpenAI was working on something competitive.

The lawsuit claims OpenAI knew about IO's technology and branding through meetings dating back to 2022. A court granted a temporary restraining order on June 22nd in IO's favor after the lawsuit was filed June 9th.

OpenAI immediately scrubbed all IO branding from their website.

Through this, though, we did learn their mystery device is NOT an in-ear wearable and might not be called IO anymore. 

What it means: 

This trademark dispute is absolutely nothing. IO versus IYO isn't even the same name.

Altman sharing those private emails was a power move to kill this lawsuit fast, or at least minimize it and turn it from an obstacle to roadbump. 

The real story is confirmation that OpenAI's hardware isn’t in-ear.

ElevenLabs dropped 11 AI, a voice assistant that executes real tasks instead of just chatting. The alpha version is live at 11.ai right now and it's pretty impressive for a first iteration. 

This isn't your typical voice assistant giving weather updates. 

(We can’t wait to ditch Siri and Alexa so we’re sooooo here for this.) 

This thing connects to actual work tools and gets stuff done. You can tell it to plan your day, update project management tools, or research prospects.

It integrates with Perplexity, Linear, Slack, Hacker News, and Google Calendar with new connections rolling out weekly.

The system is built on Anthropic's model context protocol, so it connects with Salesforce, HubSpot, Gmail, and Zapier. It uses a permissions model so you control what actions it can take within connected apps.

ElevenLabs is positioning this as direct competition to Perplexity's new iOS app and Amazon's Alexa Plus, which finally got smarter through their Anthropic Claude partnership.

What it means: 

ElevenLabs just made one of the smartest pivots in AI this year. They dominated text-to-speech and now they're using that foundation to grab actionable agent market share.

This could squash the gap between voice commands and actual productivity.

While everyone else focuses on chatbots, ElevenLabs built something that replaces human tasks. The weekly integration rollouts will determine if this becomes a legitimate productivity multiplier.

The U.S. Senate has reportedly struck a compromise to pause state AI regulation for five years in exchange for $500 million in federal AI infrastructure funding, dependent on the GOP’s ‘Big Beautiful’ budget being passed. 

Senators Marsha Blackburn and Ted Cruz negotiated this compromise, reducing the original proposed moratorium from ten years to five. It's part of the massive spending bill nicknamed the "big beautiful bill."

Here's how it works: States get access to hundreds of billions in federal AI funding, but only if they agree not to create new AI regulations for five years.

The original 10-year ban was reportedly causing rifts in the majority Republican party. This five-year compromise with (a kinda) opt-in structure gets more Republicans on board.

The Senate could vote as early as today.

What it means: 

This is the federal government bribing states to stay out of AI regulation. It's an admission that states might actually create useful AI oversight or that they could screw it all up. 

California holds all the cards here TBH. 

They can ignore federal funding and regulate AI anyway since that's where the entire industry lives. This five-year pause, if approved and enacted, gives AI companies massive runway to build without state interference.

Anthropic upgraded their Artifacts feature to let you embed Claude AI functionality directly into any app you build. This sounds cool but has major limitations.

Artifacts is their side-by-side interface where you dump data on the left and Claude builds something on the right. The new update lets you build web apps without code and embed AI technology inside those apps.

Concrete example: Upload your data, create a web app, then embed Claude inside that app so users can interact with AI that lives inside your artifact.

Google already launched this exact feature in their Gemini Canvas mode three weeks ago after their I/O conference. The Google version is completely free to use and users don't even need to be logged in to use the baked-in AI capabilities. 

What it means: 

Anthropic just copied Google's homework, but we all win. 

Side note – hit us with a reply to this email if you wanna see more coverage of this no-code embeddable feature. 

Awkard. 

Salesforce CEO Marc Benioff revealed that AI now handles 30 to 50% of the company's total workload. 

This would be massive if it were actually true. 

Benioff said AI performs tasks once done by humans including software engineering and customer service. Current employees supposedly focus on higher value work now. The company aims to have 1 billion AI agents by year end.

Their AI product reportedly reaches 93% accuracy for customer service tasks without human supervision. Benioff claims Salesforce outpaces competitors because of more extensive data and metadata.

What it means: 

Benioff seems to be exaggerating these numbers. If AI really did half of Salesforce's work, their margins would be through the roof.

This is pure marketing positioning to get attention for their AgentForce platform.

The timing coincides with them trying to compete with Microsoft and Google in enterprise AI. Real companies don't announce productivity gains this massive unless backed by financial results. 

Shots fired. 

Google launched Gemini CLI, a completely free command line interface that brings AI coding tools directly to developers' desktops. 

The tool gives developers direct access to Gemini 2.5 Pro models through the terminal. It allows 60 requests per minute and 1000 per day at zero cost. It's open source under Apache 2.0 license and available on GitHub right now.

Developers can use Gemini CLI for writing code, debugging, generating content, research, task management, and basically anything software development related.

What it means: 

This could absolutely destroy Anthropic's Claude Code line of business.

Why pay (a ton) for Claude Code when Gemini CLI is free. 

This puts massive pressure on paid coding tools. Anthropic's Claude Code isn't cheap and now faces direct competition from a free unlimited alternative using the world's most powerful model.

Cursor, Windsurf, and OpenAI's Codex tools also face new competition. The open source nature means developers can modify and extend the tool for specific needs.

This is Google's classic playbook: offer premium features free to grab market share, then monetize through other channels later.

Anthropic loses a major revenue stream here. Their coding tool business was probably generating millions monthly from developer subscriptions.

The talent wars are heating up. 

Meta reportedly hired at least eight OpenAI researchers in the last week and a half according to reports from The Information and Wall Street Journal. 

This is an all-out talent war.

This aggressive recruitment follows Meta's April launch of Llama 4 AI models that reportedly fell short of Mark Zuckerberg's expectations. Since Llama 4 flopped, Meta has been acquiring companies and poaching talent aggressively.

They spent $14 billion for 49% stake in Scale AI and got their CEO to run Meta's superintelligence team. That was brilliant because Meta gets revenue from equity AND hurts competitors who can no longer use Scale AI for data labeling and model evals. 

Sam Altman claimed Meta offered signing bonuses up to $100 million and annual salaries over $100 million. Meta CTO Andrew Bosworth says the real offers are more complex but multiple sources confirm the $100 million figures.

The researchers Meta poached reportedly helped develop OpenAI's reasoning models. These are some of the brightest minds at the company.

What it means: 

Meta is playing catch-up after Llama 4 disappointed. Throwing $100 million at individual researchers shows how desperate they are to compete.

This talent war will escalate massively.

Expect other AI companies to start matching these insane compensation packages. Meta's strategy of buying talent instead of building internally suggests they're behind technically.

Someone is breaking into OpenAI’s home. 

(It’s metaphorically Meta.) 

That’s reportedly what Mark Chen, OpenAI's chief research officer, said in a forceful memo on Slack addressing the staff exodus. The memo was leaked to Wired and expressed a sense of violation. Chen promised aggressive action to retain key researchers.

OpenAI is facing an intense talent crisis after Meta recruited eight senior researchers to join their superintelligence lab. Internal tensions are reaching a breaking point.

OpenAI leadership including Chen and Altman are working around the clock to counter competing offers. They're recalibrating comp and exploring new ways to reward top talent.

The internal memo included encouragement from other research leaders urging staff to reach out if pressured by Meta's exploding offers.

Wired's headline quoting OpenAI’s memo captured the sentiment: “someone has broken into our home."

OpenAI employees are reportedly working up to 80 hours per week. The company is shutting down for the week around July 4th, but Meta apparently isn't done poaching during their time off.

What it means:

 OpenAI is taking the talent wars seriously after losing a TON of research talent. 

When your Chief Research Officer compares competitor recruiting to a home invasion, things are gonna escalate. 

It’s all about whose got more money, the most compute and the best culture to develop AGI and beyond. 

Grab the popcorn shorties.