Dario Amodei, CEO of Anthropic, warned that security vulnerabilities discovered by the company's latest artificial intelligence (AI) model, "Claude Mythos," must be fixed within the next six months to a year.
According to CNBC and other foreign media on the 5th (local time), Amodei stated accordingly at an event co-hosted with JPMorgan Chase. He said China's AI models are about six to 12 months behind Mythos, and stressed that the problems Mythos found must be resolved within that window.
Mythos is the latest AI model that Anthropic released in Apr., with strengths in detecting security vulnerabilities in operating systems (OS) and web browsers. However, concerns have also been raised that if the vulnerabilities the model discovers are exploited in real attacks, they could pose serious threats to national security, financial systems, and critical infrastructure.
Amodei said, "It will be a major threat if the number of vulnerabilities and breaches increases and the financial damage from ransomware grows across various institutions such as schools, hospitals, and banks." He also warned that if the vulnerabilities identified by Mythos are disclosed, "malicious attackers will exploit them."
Amodei, however, said there is still time to respond to the coming cybersecurity threats. "It's a dangerous moment, but if we respond correctly, we can create a safer world afterward," he said, adding, "The bugs we need to find are limited, and we have already begun to take the first steps."
On AI regulation, he argued that a balance is needed between safety and industry growth. He said, "We need to find a system that allows the industry to operate quickly and fairly while putting safeguards in place for serious problems."