The AI That Chose Rules Over Speed: What Is Anthropic and Why Does It Matter
At the recent Artificial Intelligence Summit held in New Delhi, 86 countries signed the most ambitious diplomatic agreement on this technology to date. But one of the most revealing moments of the event was not a document or a speech: it was two men who decided not to shake hands for the photo. Sam Altman, CEO of OpenAI, and Dario Amodei, founder of Anthropic. Two sides of the same coin, or perhaps two completely different coins.
From the Same Table to Rival Companies
OpenAI was founded in 2015 and quickly began developing the GPT models we know today. During that process, Amodei, then VP of Research, built up a specific frustration: he felt the company was prioritizing speed to market over truly understanding the risks of what they were building. With technology that new, those risks were even less understood than they are now.
That disagreement did not stay behind closed doors. A dozen high-profile engineers left alongside Amodei and founded Anthropic, with a different premise from day one: safety is not an optional feature, it is the starting point.
A Constitution for AI
While OpenAI and other models were trained using reinforcement learning from human feedback, Anthropic developed its own approach: Constitutional AI. The idea is straightforward: the model operates with a defined set of rules and values that it considers in every interaction. It does not only learn from what humans approve of, but from principles embedded at the design level.
Their language model is called Claude and can write code, generate software, and build applications. One of its more recent versions allows users to automate tasks directly from their desktop: organizing files, managing emails, planning the workday, all from a single instruction.
The Cost of Being Trustworthy
Anthropic is not chasing millions of free users. Many of its key features require payment, and that is a deliberate choice. That commitment to safety made it the only AI platform operating within classified environments of the United States government.
However, that position also put them in a difficult spot. Reports from specialized outlets indicated that Claude was used by the U.S. military in the operation related to the capture of former Venezuelan president Nicolás Maduro. Anthropic demanded explanations from the Department of Defense, as the company has consistently asked to be informed about how and why its technology is being used. The government's response was the opposite: pressure to gain unrestricted access to all models, with no requirement to report each use.
On February 22nd, Anthropic modified its Constitutional AI. Previously, the company required that a model's safety be demonstrated before public release. That condition no longer appears in the updated version. Several experts suggest that pressure from Washington may be a factor.
A Tension That Is Not Going Away
From OpenAI and models like Grok, the criticism toward Anthropic is that the company wants to impose its own vision of how AI should be used, slowing down progress. Anthropic's position has been consistent: opposing the automation of processes involving weapons is not a matter of protocol, it is a matter of conviction.
The irony is hard to miss. The people who left OpenAI precisely because they refused to compromise on their principles are now facing institutional pressure that puts them in the exact same uncomfortable position. The race for artificial intelligence is not only technological. It is also philosophical.