Photo: internetua
The CEO of Anthropic, Dario Amodei, and Jamie Dimon discussed growing cyber risks linked to artificial intelligence, highlighting how rapidly AI systems can discover security flaws in modern software.
Amodei said that advanced AI models can already detect vulnerabilities faster than companies are able to patch them, creating a narrow window—estimated at 6 to 12 months—to fix tens of thousands of issues before similar tools become widely accessible to attackers.
He warned that the biggest risk is a rapid increase in exploitable bugs, data leaks, and ransomware attacks affecting critical sectors such as healthcare, education, and finance. According to him, Anthropic avoids publicly disclosing discovered vulnerabilities until they are fixed, to prevent misuse.
The discussion referenced internal tools like “Mythos,” which reportedly identified hundreds of vulnerabilities in controlled tests, including a large-scale audit of the Firefox browser in collaboration with Mozilla. Some government agencies are also said to be experimenting with Claude-based systems for security scanning of restricted networks.
Industry reactions remain divided. OpenAI CEO Sam Altman has previously criticized such warnings as exaggerated, while Amodei argues that regulation should apply evenly across all AI developers and that the current period is a critical “window of opportunity” to strengthen cybersecurity before these capabilities become widespread.
