Jornal Do Brasilia - Anthropic says will put AI risks 'on the table' with Mythos model

"We have a model that's beginning to outstrip human capabilities in the cyber world," Anthropic's Paris-based chief of relations with startups and tech firms Guillaume Princen told AFP in an interview.

Mythos is "capable of spotting security holes that have existed for decades, in systems tested by both human experts and automated tools, that have never been discovered before," he added.

Anthropic has delayed a general release of Mythos, sharing it first with a few dozen key American tech and financial services players -- such as Nvidia, Amazon, Apple and JP Morgan Chase -- to allow them to test and improve their security infrastructure.

But the company has also been accused of overhyping the powers of a technology which is its stock in trade -- and the subject of fierce competition with rival OpenAI.

The Mythos news broke as rumours grow that Anthropic will list on the stock market this year.

- Safety first? -

"We prefer to be transparent and lay these risks out on the table," Princen said, adding that AI safety concerns are "central to Anthropic's DNA".

"We don't have all the answers, this has to be a conversation between tech actors like us who have the data, the academic world, the political world and the world of economists," he added.

Mythos' reported capabilities have unsettled the American financial sector and the European Union, which requested more information from Anthropic.

In an open letter to businesses, the British government said that Mythos "highlights the speed at which AI capabilities are increasing and the threats they potentially pose".

No European company is part of Anthropic's "Project Glasswing" consortium for shoring up cyber defences using Mythos' findings.

That has raised questions about how prepared the rest of the world will be for the offensive capabilities of US-owned AI.

Mythos is "certainly not a model that will soon be opened to the public at large, for obvious reasons," Princen said.

Anthropic is nevertheless "thinking about the next waves of opening up," he added.

- European growth -

Europe is the region where Anthropic sees the fastest growth.

Its Claude Code software development tool generates around $2.5 billion in annualised revenue -- a figure based on extrapolating from a few recent weeks of sales.

Much of that expansion comes from "European firms riding the wave" of AI, Princen said.

The company has opened offices in Dublin, London, Paris and Munich, and wants to keep investing across the continent.

"We go where the demand is," Princen said, pointing to partnerships with European firms like Swedish coding startup Lovable or Danish pharma company Novo Nordisk.

Relatively unknown to the wider public until recently, Anthropic was founded in 2021 by former OpenAI staff and makes around 80 percent of its revenue from business-to-business sales.

The company and its Claude chatbot surged in prominence in late February, when bosses refused to allow its AI tools to be used by the Pentagon for mass surveillance of American citizens or fully autonomous weapons.

The Trump administration responded by designating Anthropic a so-called "supply chain risk" to national security -- a decision being contested in multiple legal cases.

In legal documents seen by AFP, Anthropic finance chief Krishna Rao warned that Washington's move could cost the firm multiple billions in revenue this year.

On the other hand, "there are a lot of people who started using Claude precisely because of the position we took on that question," Princen said.

Anthropic said in early April that it had tripled its annualised revenues quarter-on-quarter to over $30 billion -- outpacing OpenAI for the first time.

F. Tavares--JDB