ARTIFICIAL INTELLIGENCER-Why Anthropic is turning away some users

By Krystal Hu

Oct 21 (Reuters) - Bill Gross, founder of Idealab, is Silicon Valley’s idea guy.

Over 30 years, he’s developed a methodology for spotting business ideas and used it to help start 175 companies through his incubator, with more than 45 exits.

So it stung a little when, a few months ago, Gross found that AI had gotten disturbingly good at generating business ideas. He told me it left him feeling “depressed”. Then the feeling turned into an idea itself: what if he turned his methodology into a system anyone could use?

That system, Idealab.ai, launched this week. It matches your background and skills against AI-generated business ideas, and lets you set the budget you’re willing to work with.

Not every idea it spits out is going to be a great business, but Gross is betting the platform can lower the barrier to entrepreneurship for people who’d never otherwise know where to start.

The human part, he insists, doesn’t go away. “All the judgment and decisions still have to be made by humans,” he told me. AI can offer suggestions, “but you have to make the call.” And, “you’ve got to be passionate about it, for sure.”

OUR LATEST REPORTING IN TECH AND AI

Musk says SpaceX-Tesla merger is 'a good idea' as companies form deal committees

Trump says companies should pay 'dividend' to communities to make data centers popular

OpenAI says its AI agents weakened internal monitoring system

Qualcomm to cut jobs worldwide in performance-focused layoffs, internal memo shows

SpaceX and Tesla's Terafab targets 2028 for volume chip production, Musk says

ByteDance, Alibaba, Tencent place over $35 billion worth of Nvidia chip orders

ANTHROPIC'S FIGHT WITH DISTILLATION

Here’s a strange move for a company racing toward the largest IPO in history: turn away paying customers, on purpose, weeks before you go public. That’s effectively what Anthropic just did.

Inside an investor meeting in San Francisco last week, my colleagues and I got a behind-the-scenes look at how Anthropic’s executives are explaining a slowdown of their own making.

The company has cut off a meaningful slice of paying customers — accounting for between 5% and 7% of its revenue in the first nine months of the year, investors at the meeting told us. The reasoning: those accounts weren’t typical customers. They were, Anthropic alleged, Chinese AI labs quietly mining Claude to train their own models.

The technique is known as distillation in AI model development. A smaller “student” model learns by studying the outputs of a larger “teacher” model, essentially absorbing its capabilities secondhand. In February, Anthropic said Chinese labs including DeepSeek, Moonshot, and MiniMax had generated more than 16 million interactions with Claude through roughly 24,000 fake accounts, all in violation of its terms of service.

Anthropic calls this more than a contract violation. It considers the practice a loss of intellectual property with national security implications. So the company rolled out new defenses earlier this month: applying access restrictions, proxy blocking, and bans on tens of thousands of accounts, which it says has cut off 90% to 99% of the data being siphoned.

Anthropic is telling investors the crackdown has effectively stopped Chinese developers from distilling its models.

Here’s why that matters for the IPO: it gives Anthropic a tidy explanation for something investors were already nervous about — a month-to-month growth deceleration. It may also address a bigger worry. Anthropic is betting that cutting off Chinese labs will slow their progress, not just stop a few bad actors from gaming the terms of service.

The underlying numbers are still extraordinary by any standard. By this past March, Anthropic had already brought in $4.7 billion in a single quarter — more than all of 2025. Then it kept climbing: $11.6 billion in the second quarter, $16.8 billion in the third.

But growth at that scale doesn’t usually keep compounding at the same rate forever. Competition from cheaper, open-source Chinese models remains one of the most common concerns investors are raising.

Reuters has reported that some investors are now discussing a valuation near $1.5 trillion for the IPO, down from the roughly $2 trillion once expected. The real question is whether this crackdown — and the short-term revenue hit that comes with it — actually buys Anthropic a lasting distance from its open-source rivals, or just a temporary edge.

CHART OF THE WEEK

https://www.reuters.com/graphics/BRV-BRV/byprabxyave/chart.png

According to a new survey of more than 380 security and technology professionals by Salt Security, 86% of their organizations now have at least one AI agent operating, and 77% have agents chaining actions together across other agents. In other words, agents aren’t just answering questions anymore — they’re taking multi-step actions, often autonomously, deep inside company systems.

What’s missing is the ability to actually watch them. Only 18% of those same organizations say they can automatically stop an agent that starts behaving unexpectedly, and just 13% can trace a complete path from instruction to outcome across every agent involved. That’s a massive visibility gap between how fast companies are deploying agents and how well they can monitor what those agents are actually doing.

(Reporting by Krystal Hu, additional reporting by Milana Vinn and Echo Wang, Editing by Rosalba O'Brien)