Dario Amodei issued a warning on Tuesday that deserves to be taken seriously.
At Anthropic’s financial services event in New York — standing alongside Jamie Dimon on stage — the Anthropic CEO explained that AI has created a narrow time window. Mythos, Anthropic’s most powerful model, has found tens of thousands of vulnerabilities in common software. The question isn’t whether these flaws will be exploited — it’s when.
The numbers
An earlier Anthropic model found roughly 20 security vulnerabilities in the Firefox browser. Mythos found nearly 300. Across all tested software, the total count runs into the tens of thousands.
The problem: China’s AI models are only six to twelve months behind Mythos, according to Amodei. That means the same vulnerabilities will soon be discoverable by models that don’t operate under Anthropic’s controlled access conditions.
‘Moment of danger’
Amodei called the current situation a ‘moment of danger.’ His concern: a massive increase in security breaches, ransomware attacks on schools and hospitals, and financial damage on a large scale.
But he didn’t stop at warnings. ‘If we respond to it correctly — and I think we’ve started to take the first steps — then we can have a better world on the other side,’ he said.
What Anthropic is doing
Anthropic keeps Mythos under lock and key — only selected partners like Goldman Sachs and JPMorgan have access. Claude Security recently entered public beta, offering automated vulnerability scanning for enterprise customers.
That’s the core dilemma: Mythos is simultaneously the most powerful defense tool and the most dangerous weapon in the wrong hands.
My take
Six to twelve months. That’s not a lot of time when you consider how slowly large organizations patch their systems. Amodei’s warning isn’t marketing — it’s a real countdown.
Sources: