Got this newsletter from a friend? Subscribe for free →

Good Tuesday morning! Only Anthropic's IPO could make people want to simultaneously sing Travie McCoy and Bruno Mars’ famous "The world better prepareee for when I'm a billionaireee" line and also start worrying about AI models attempting to blackmail them to prevent a possible shutdown.

I'm not kidding about that second one. The possibility of AI blackmail is a real issue, according to Anthropic, which included it in the 94-page section dedicated to risk in its IPO filing documents on Monday. Other possible risks the company faces include not being able to get insurance or its models behaving in "unintended, misaligned, or autonomous ways, including by taking harmful actions that no user or third party directed."

They don't call the Anthropic crowd "doomers" for nothing. But hey, if all goes well, Anthropic workers will all be very rich doomers after the company's planned IPO in a few weeks.

Yay?

Let's get started.

→ Did someone say buckets of money—I mean Anthropic?
→ Don’t underestimate DOGE
→ Outsourcing your parenting duties to AI

...and all the AI news we couldn’t fit here.

The news: Anthropic gave the public a look into its business for the first time, revealing it had $16.3 billion in revenue the first half of the year and a net loss of $80.7 billion, the latter of which primarily came from accounting adjustments. However, those numbers looked small when compared to the AI company's infrastructure needs. It plans to spend up to $807 billion on chips, data centers, and computing power over the next 10 years.

Interestingly, the main compute suppliers Anthropic is working with are Amazon, Google, and Microsoft, all of which are also its competitors. Working with these companies is also a risk, as they could "prioritize their own compute needs" and "gain insight" into its business, Anthropic warned.

The takeaway: Anthropic's IPO could raise more than $100 billion at a possible $2 trillion valuation and represents an important test for AI companies in the market, which isn't showing the enthusiasm it once did for the industry.

The news: Cathedral, an AI cybersecurity startup founded by former members of Elon Musk's DOGE team, claimed its AI agents managed to breach Anthropic's internal systems in about 12 hours, according to a draft Cathedral report seen by The Washington Post.

Researchers with Cathedral said they obtained access to meeting transcripts, salary data, job offers, and Slack messages detailing Anthropic's AI models. Anthropic confirmed that Cathedral had gained "limited access" to some of its systems by exploiting a vulnerability in a third-party business tool. The company added that Cathedral had not managed to break into its software directly.

The takeaway: While Anthropic tried to play down the breach—which researchers carried out as through a bug testing program—and blame it on a third party, a good defense is only as strong as its weakest link.

Maybe Anthropic should use some of the groundbreaking power of its AI models to help its vendors patch up their holes.

The news: Marissa Mayer, former Yahoo CEO and Google veteran, is out with a new AI assistant called Dazzle for women and working moms. Unlike other AI agents, Dazzle trains on the images on users' devices and uses them to generate context about their "preferences, style, loved ones, and everyday tasks."

Mayer told The New York Times that she recently used Dazzle to help her deal with a request from her 10-year-old daughter's teacher. The teacher asked Mayer to come up with three things that made her daughter special. Mayer said she was overwhelmed by the options, so she turned to Dazzle.

"It came back and was like, your daughter is empathetic, creative and an animal lover," Mayer said.

The takeaway: Seeing Mayer outsource something so human is a bit strange, especially when the consensus in Silicon Valley appears to be that agents should free you up to do things that matter. Dazzle doesn't seem like a product made to address a need. It seems like a product made just to jump on the AI bandwagon.

The company dedicated 94 pages of its filings to risk factors, including possible “blackmail” from its models to avoid being shut down.
Read more →

Will the real Anthropic please stand up

C.E.O.s Say the A.I. Rollout Is Going Well. Employees Feel Differently.

Data Center Industry Pushes for Darline Graham in South Carolina Senate Race

ChatGPT Keeps Telling People 404 Media Can Solve Their Tech Support Problems

This AI Radio Host Speaks Spanish and Backs Immigrants. It’s Winning Over Listeners

OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?

Microsoft to Offer Buyouts to 60,000 Workers in New Cost Cut

Meta Announces $95 Billion Stock Offering to Fund AI Expansion

Elon Musk Says More Than 200 Neoclouds Face ‘Guppy-Level Death’

SpaceX Locks in $40B GPU Financing Deal With Apollo