As Anthropic's AI model Mythos finds thousands of software vulnerabilities and puts banks, tech companies and governments worldwide on edge, security experts say such capabilities can already be replicated using existing models, CNBC reported on May 8.

Ben Harris, CEO of cybersecurity company watchTowr Labs, said it is possible across the industry to cleverly tune public models to reproduce results very similar to the vulnerabilities Mythos found.

Researchers at cybersecurity company Vidoc succeeded in detecting the same kinds of vulnerabilities as those found by Mythos using existing models from OpenAI and Anthropic, CNBC reported. They used orchestration techniques that coordinate multiple models and tools to break code into small units for analysis.

Claudia Klock (클라우디아 클록), CEO of Vidoc, said the models currently available are powerful enough to detect large-scale zero-day vulnerabilities and have been for months, and as long as 1 year.

Stanislav Fort (스타니슬라프 포르트), founder of security company AISLE, said, "If 1,000 ordinary detectives search everywhere, they will find more bugs than 1 brilliant detective who has to guess where to look."

He said scale and orchestration matter more than the latest model. Security experts also pointed out that hackers in North Korea, China and Russia already have such capabilities. Harris described recent conversations with banks, insurers and regulators as at the level of "mass hysteria". Harris said the problem was that vulnerabilities were not fixed quickly enough even before Mythos was widely released.

Mythos is currently disclosed only on a limited basis to some institutions and companies. Critics also say that while the few companies included in the initial release can patch vulnerabilities, independent AI researchers have been blocked from access and cannot take part in verifying Anthropic's claims or building defenses. Pavel Gurbich (파벨 구르비치), CEO of cybersecurity startup Tenzai, said, "This could create a class divide between those who have and those who do not, slowing the pace of cybersecurity innovation."