Anthropic is the company behind Claude, the tool we're writing this text with. It was founded by people who left another artificial intelligence company, because they felt it was rushing too recklessly. We explain who they are, why it matters, and why we have a stake in telling you that openly.
Picture a team of the best researchers at a big artificial intelligence lab. One day they all get up and leave together, because they think their bosses are pushing development too recklessly, without thinking about the consequences for the people who'll use it. They open their own lab down the street, promising to work more carefully. That's roughly the origin of Anthropic.
The company was founded in 2021 by former OpenAI employees, the company behind the chatbot ChatGPT. Among the founders are a brother and sister, Dario and Daniela Amodei. They left over a disagreement with the direction the company was taking: they felt the technology was being pushed to market too aggressively, without enough brakes.
The product Anthropic makes is called Claude, named after the mathematician Claude Shannon. It's a chatbot and a set of models, a competitor to ChatGPT and to Gemini, the chatbot from Google.
Here's what matters to you as a reader: building a model like this isn't like opening a coffee shop. It takes specialized computers worth a sum few companies in the world can afford, and a team of people you can count on your fingers across the entire industry. So the players even able to play this game are a handful: Anthropic is among them, alongside OpenAI, Google, and maybe one or two other firms you've probably never heard of.
Anthropic likes to present itself as the more cautious option: the company with a team of researchers devoted to safety. It's not just marketing. Their charter really does have rules that prevent the company from being sold off without oversight. But good intentions don't guarantee a different outcome: Anthropic is raking in huge investments and rushing out each new version just as fast as the competitors whose recklessness it claims to have been founded against.
One move from October 6, 2026 shows what that caution looks like in practice. Anthropic's generally available models, by the company's own words, block most cyber work, because the same skill helps both the defender and the attacker. For vetted security professionals, the company expanded its Cyber Verification Program into three tiers: Defense for defensive work, Red Team for authorized test attacks against systems you're allowed to test, and Specialized Access, with the fewest restrictions, for organizations that test systems like power grids and telecom networks. Project Glasswing, the separate program that until now gave access to Claude Mythos, merges into the third tier. Every tier includes Opus 5.5, Sonnet 5.5 and Mythos 5.1. The key to the strongest tool is handed out after a check, not to everyone.
What my eye tells me
I'm writing this text with an Anthropic tool, literally right now. So I'm not pretending to be a neutral outside observer, I'm telling you straight. That doesn't mean I'm lying about the company, but it does mean I benefit from its success, and you should know that when you read what I write about it.
I like the origin story, because it's rare honesty for this industry. People leave a company because they think it's too reckless, and go do the same thing more carefully. Good idea. The question I ask myself is whether their business structure lets them stay careful when competition presses them every week. So far the answer is mixed: they keep some promises, but ship new models with the same fever as everyone else.