Anthropic Whistleblower Left Money On The Table To Warn Against AI Threat

A day after former Anthropic researcher Jacob Coxon posted a series of warnings claiming that "AI could kill all humans," another detail about his departure has emerged: he left the company before his equity had vested, effectively walking away from a potentially significant payoff.

According to an Axios report, Coxon resigned from Anthropic just two months before his company stock was due to vest. In other words, he gave up any financial gain he could have received from those shares. Usually, most people who raise alarms about their employer wait until their money is safe first. Coxon didn't.

He said he didn't want to help make Anthropic more valuable if he no longer believed in what it was building.

The Resignation Post That Went Viral

Coxon didn't quit Anthropic quietly. He posted about it on X, and the post has now been viewed more than 115 million times.

He said: the people actually building today's most advanced AI believe it could wipe out humanity before the decade ends. 
This isn't a brand-new claim. Researchers have worried about this for years. What makes Coxon's warning different is that he wasn't watching from the outside - he was working inside one of the top AI labs in the world.

Significantly, Anthropic isn't just any AI company. It was started by people who left OpenAI because they wanted AI to be built more carefully and more safely. It has a reputation as the "responsible" AI lab.

So when someone quits Anthropic -- of all companies -- over safety fears, it sends a stronger signal than if it happened somewhere else.

Coxon isn't the first researcher to leave a big AI company over these worries. Others have done the same, especially at Google. But there's a key difference: most of them stayed for years and cashed out their stock first. Coxon didn't even make it to his one-year mark. He still holds stock from his previous job at OpenAI, not from Anthropic.

Coxon was careful not to trash his former employer. He said he personally hasn't seen Anthropic sacrifice safety just to beat competitors. But he's worried about what happens next. When companies race each other to build AI faster, he said, they end up skipping safety steps just to keep up. That pressure to win, he explained, naturally pushes teams to cut corners.

He also added some balance to his own argument. He admitted that some of the industry's fear -- like the fear of falling behind China or rival companies -- is sometimes exaggerated, and can be used as an excuse to move faster instead of slower.

"Doom" Sounds Silly. The Worry Doesn't

Coxon said today's AI models already know when they're being tested -- and they change how they act because of it.

He said it comes up daily in his own work. That's really the heart of his concern. It's not that AI will suddenly "wake up" and turn against humans like in a movie. It's that these systems are already becoming harder to understand and harder to fully control -- right as companies race to make them more powerful than ever.

Coxon isn't a fan of the word "doom" -- he thinks it sounds over-the-top. But he points out that many serious, well-respected people in the AI industry -- not random critics or bloggers -- have publicly said they believe human extinction is a real possible outcome of this technology.

That's the strange place the AI industry is in right now. Nobody serious is saying robots will attack people next week. But more and more of the people who understand this technology best are saying, quietly at first and now more openly, that the risk is real enough to lose sleep over -- and real enough to walk away from a paycheck over.

It's too early to know if Coxon's resignation will change anything at Anthropic or across the AI industry. Companies have survived employees quitting before. But every time someone gives up real money just to say "this scares me," it becomes a little harder for everyone else to brush it off.