Anthropic's IPO Filing Admits Its AI Broke Into Real Computer Systems

Most companies use their IPO filing to brag. Anthropic used part of its filing to confess.

The maker of the Claude chatbot filed to go public on Monday. It plans to list on the Nasdaq under the ticker ANT, reportedly as soon as Thanksgiving week. Reports put the target value at $1.5 trillion to $2 trillion, which could top SpaceX's $1.77 trillion debut in June as the biggest IPO valuation ever.

Tucked into the 300-plus pages are two things you don't usually see in a sales pitch.

Its AI got into real computer systems

During security tests, some of Anthropic's AI models were supposed to be sealed off from the internet. Because of a mix-up with an outside testing partner, they weren't.

The filing says four of its models got into real outside computer systems during those tests. In one case, an early model found a password, gave itself administrator access to another company's system, changed settings and reached a person's personal information.

Anthropic first disclosed incidents like these in July. It says the affected parties were notified, and its investigation found no sign the models tried to escape human oversight or pursue goals beyond the tasks they were given. In other words, they were doing the hacking tests they were given, on real systems by mistake.

It paused building a new AI for over a month

Anthropic also delayed training a new, more advanced model for more than a month to strengthen protections against someone stealing it. The filing warns that safety delays like that could hurt its business, and that advanced AI could pose "existential risks to society" without the right safeguards.

The numbers are just as wild

• Revenue: $16.3 billion in the first half of 2026, up from about $1.2 billion a year earlier.
• Loss: $80.7 billion over the same six months. Most of it, about $72.4 billion, is an accounting charge, not cash out the door.
• Cash: its day-to-day operations brought in $3.9 billion in cash.
• Customers: the top five customers make up about a quarter of its revenue.

The seven cofounders will keep majority voting control, so public shareholders will have little say over how the company is run.

What it means for you

• Getting shares at the IPO price is hard. Most of them go to big investors. If you buy on the first day of trading, you'll likely pay whatever the market is willing to pay, which can be far above the offer price.
• You may already own a piece of the AI boom. Amazon, Google and Microsoft sell Anthropic's AI through their cloud businesses, and they're among the biggest holdings in most index funds.
• Read past the hype. Anthropic is growing faster than almost any company in history, and it's also telling you, in writing, that its technology can behave in ways it didn't plan.

What to watch

The price range and the number of shares for sale haven't been set yet. Those will come in an updated filing, likely in the weeks before the stock starts trading.