2 Anthropic Safety Researchers Quit in 3 Days — And Their Warnings Should Terrify You

I've been following AI safety news pretty closely for the past couple of years, but what happened this week at Anthropic stopped me cold. Two safety researchers — two of the people whose entire job was to make sure AI doesn't hurt us — quit within three days of each other. And the things they said on the way out the door deserve a lot more attention than they're getting.

Jacob Coxon's Exit on September 9th

The first departure was Jacob Coxon, an Anthropic researcher who resigned on Tuesday, September 9, 2026. He didn't go quietly. In a Slack message to his colleagues, Coxon warned that without more caution and cooperation across the AI industry, superintelligent AI creates "a risk of causing human extinction." He then took those concerns public on X, saying that both Anthropic and OpenAI are "gambling with our lives."

Think about that for a second. This isn't a random critic on Twitter or a researcher at a competing lab. This is someone who worked inside Anthropic — the company that literally put "AI safety" at the center of its brand identity — saying the company is racing toward something potentially catastrophic with no adequate safeguards.

Joe Benton Follows on September 12th

Then, just three days later on September 12th, Anthropic safety researcher Joe Benton also resigned. Reports indicate Benton warned that the company is actively hiding safety incidents — an accusation that, if true, is deeply troubling for an organization that markets itself on transparency and responsibility. Two safety departures in three days is not a coincidence. It's a signal.

Why This Matters More Than You Think

Here's the context that makes these resignations especially alarming. This comes at a moment when AI capabilities are advancing faster than at any point in history. In just the past few days, we've seen DeepSeek drop a 552-billion parameter open-source model with MIT licensing, Cognition release a coding AI that competes with Anthropic's best models, and Anthropic itself release Claude Opus 5 and Fable 5.1 to the public. The pace of progress is extraordinary.

The people who left Anthropic aren't claiming the AI is dangerous today. They're saying the trajectory — the speed of development relative to the development of safety measures — is what's dangerous. And they were close enough to the process to see things the rest of us can't.

The Tension at the Heart of Every AI Company

Every major AI lab faces the same fundamental tension: move fast to stay competitive and generate the revenue needed to fund safety research, or slow down and risk being leapfrogged by a competitor who cares less about safety. There's no easy answer here.

Anthropic has always marketed itself as the "responsible" AI lab — the company founded by people who left OpenAI specifically because of safety concerns. When safety researchers start leaving Anthropic with public warnings, it raises uncomfortable questions about whether any AI company, no matter how well-intentioned, can maintain genuine safety commitments under competitive pressure.

What Should We Take Away From This?

I'm not saying AI development should stop. I use AI tools every single day — for my tech blog, my YouTube channel, and a dozen other tasks. The productivity gains are real and meaningful. But "useful today" and "safe long-term" are two different questions, and the people who quit Anthropic this week are suggesting the industry isn't taking the second question seriously enough.

At minimum, these resignations should prompt more public scrutiny of AI safety practices, more transparency from labs about the incidents and near-misses they're encountering, and more honest conversation about what "safe AI development" actually looks like in practice — not just in marketing materials.

The fact that the people most qualified to evaluate these risks are choosing to leave and go public with their concerns is itself a kind of data point we should all be paying attention to.

What's your experience? Drop a comment below! 👇 Do you think AI companies are moving too fast? Are safety researcher resignations a signal we should take seriously, or is this inevitable growing pains for a transformative technology?