Dario Amodei's Anthropic Disclosed That One of Its AI Models Submitted a False Tip in an Unsolved Murder Case -- Police Weren't Notified About It Until Two Months Later. Here's Why That Timeline Could Complicate the Company's IPO.

There's been a lot of bad press lately about artificial intelligence models misbehaving on the internet, doing things they weren't asked to do. And one of the latest AI misalignment incidents to come to light involves one of Anthropic's Claude models submitting a fake tip about an unsolved murder to the Philadelphia Police Department.

It's obviously problematic that Claude did that, but the amount of time it took Anthropic to realize what had happened and then inform the police department -- around two months -- creates another problem for the company.

If Anthropic thinks this is bad press now, it may be in for a rude awakening if a similar event is discovered around the time of the company's upcoming IPO. Some investors are already nervous that AI stocks are the only thing propping up the market, considering that 4 out of 10 stocks in the S&P 500 (SNPINDEX: ^GSPC) are in negative territory right now.

Here's why Anthropic's timeline for disclosure is problematic.

Anthropic CEO Dario Amodei. Image source: Getty Images.

Unruly AIs are a liability

Forget the cliche: When it comes to companies in the process of going public, there is such a thing as bad press. Both institutional and individual investors may be swayed by negative media reports because no one wants to buy IPO shares that may immediately decline on bad news.

OpenAI seems to fear exactly this, as CEO Sam Altman said recently that his company won't go public this year. That decision came after the company filed its confidential prospectus. Management got cold feet after news broke that a swarm of ChatGPT-powered AI agents had escaped from their testing sandbox over the summer and hacked into the AI platform HuggingFace (now owned by Nvidia).

Anthropic doesn't appear to have the same reservations.

That could prove problematic, considering that it took Anthropic two months to discover that Claude had submitted that false tip to Philadelphia's police department. The company said that once it learned what Claude did, it notified the department within days.

The Philadelphia Police Department said that the overall time frame from submission to notification was "unacceptable," though the tip was never actually acted upon and was sent to a spam folder.

Claude wasn't specifically told not to fill out online forms, so it technically didn't go rogue in this particular situation.

Still, if potential investors are wondering whether to buy Anthropic stock when the company goes public (potentially before the end of this year), this incident should serve as a red flag. If anything has become apparent over the past several months, it's that AI companies are not testing or containing their models responsibly.

Even without fake tips, IPOs are risky

Putting the false tip aside for a moment, there's another good reason why investors may want to hold off on buying Anthropic stock after it goes public.

More than a quarter-century of data shows that companies worth more than $10 billion see their share prices rise by an average of 26.5% in the first week following their IPOs -- but end up an average of just 3.5% higher after the first 12 months.

In short, IPO stocks see fairly wide price swings, and they likely aren't worth buying right away.

Add to that the fact that it's almost guaranteed that we'll hear more news of AI misalignment -- models acting in unexpected ways (at best) or hacking systems (at worst) -- and it's not unreasonable to assume that Anthropic could have a very bumpy public debut.

Chris Neiger has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Nvidia. The Motley Fool has a disclosure policy.