Dario Amodei, chief executive officer of Anthropic, has issued a stark warning about the future of artificial intelligence: some AI models will become so dangerous they should require a weapons license for their use. The statement, reported in an interview with Bloomberg and previewed by xataka.com, reflects the increasingly vigilant stance of this company on technological safety, just as the two major AI powers—Anthropic and OpenAI—prepare for stock market listings at astronomical valuations.

The model Anthropic chose not to release

Amodei has revealed that his company deliberately withheld Mythos, an AI model specializing in cybersecurity, because it possesses capabilities beyond what he considers safe for the general market. This system is capable of autonomously executing an entire cyberattack process, identifying exploitable vulnerabilities without human intervention.

Companies that tested Mythos in early trials were explicit in their requests: they directly asked Amodei not to release it. "Mythos is a total weapon. You should need a weapons license to use it," the executive acknowledges. Despite the negative commercial implications of this decision, Amodei argues that releasing it before most infrastructure is patched would be exponentially more dangerous.

The dilemma between values and business relevance

In the same interview, Amodei reflects on a fundamental conflict facing any major technology company: the tension between maintaining ethical principles and remaining competitive. When asked why Anthropic focuses on enterprise tools like Claude Code rather than pursuing more lucrative mass-market applications, he was blunt: "Either you betray your values or you become irrelevant".

According to his analysis, choosing a business model that contradicts a company's founding principles inevitably creates that dilemma. Anthropic has chosen the more demanding path.

The quiet threat to office employment

Beyond technical matters, Amodei has been warning for over a year about a troubling economic scenario: the possibility of surging GDP growth coupled with very high unemployment, particularly in entry-level office jobs. His forecast is alarming: AI could eliminate half of these positions within one to five years.

He has faced criticism from both sides. Nvidia CEO Jensen Huang accuses him of confusing tasks with jobs. Others point to alleged "apocalypse marketing" that would commercially benefit Anthropic. Amodei responds by emphasizing that his message has never been purely catastrophic, but rather a call for political action: he proposes taxes on token usage and joint workforce redesign with companies.

The mechanism he describes is simple but worrying: when 90% of a task is automated, the remaining 10% becomes much more productive. The danger emerges when automation approaches 100% and nothing remains to hold onto. Within Anthropic they are already experiencing this transition; AI writes virtually all code, though for now it remains a productivity multiplier rather than a replacement.

The context of the AI gold rush

These statements come at a time when Anthropic and OpenAI are valued at astronomical figures—around 965 billion and one trillion dollars respectively—as they prepare for stock market debuts that would mark a historical milestone. In this context, Amodei's insistence on restraining dangerous capabilities becomes even more significant, as it represents a real economic cost in favor of caution.

The lingering question is whether other AI companies will follow his example or whether competitive pressure will eventually erode these ethical barriers.