Companies that have tried Anthropic’s AI model Mythos, which it first made available to some users for cybersecurity, have generally responded positively. Some say it not only found software security vulnerabilities compared with earlier efforts, but also uncovered flaws that had gone undetected for a long time.

Sam Rubin (샘 루빈), a senior vice president who leads the threat intelligence team at global security company Palo Alto Networks, said Mythos found more than 20 vulnerabilities in 3 weeks. Several other companies, in addition to Palo Alto Networks, have shared similar experiences with Mythos.

Discussion around Mythos has so far focused on which vulnerabilities it found and how quickly. Whether the process involved significant costs has largely remained unclear. But recent reporting by The Information suggests that companies without deep pockets may find it far from easy to use Mythos, even if they can access it.

The report said Palo Alto Networks burned through about $1 million worth of tokens very quickly while using Mythos. Another company that tested Mythos also shared that it could use millions of dollars’ worth of tokens in a matter of weeks.

Because Anthropic has been providing subsidies to early Mythos test users, Palo Alto Networks did not need to pay testing costs. But the situation could change once the test programme ends.

The Information reported that Anthropic has said Mythos would cost about six times more per token than Opus, the company’s top-performing model currently on offer. Opus is also often described as expensive in the AI market. If Mythos is about six times more expensive than that, it could mean a sizeable cost burden for users.

Even so, many assessments say Mythos is worth paying for because it performs very strongly on cybersecurity tasks compared with Opus.

The Information cited a British organisation that tested Anthropic models as saying token prices may be six times higher, but when factoring in work efficiency, customers may end up paying only about twice as much as Opus to use Mythos.

Global security company Zscaler also says Mythos is indeed much more expensive than existing code-scanning tools, but still worth the investment. Zscaler has been using Mythos to find vulnerabilities in its product code.

Other corporate security leaders say paying for Mythos is necessary, given that major breaches can cost hundreds of millions of dollars when legal costs and compensation for damages are included. The Information reported that corporate cybersecurity executives are preparing to increase budgets for spending on Mythos and other advanced AI.

They are also conducting internal assessments on whether they should buy additional cybersecurity tools to counter attackers using AI to penetrate corporate systems.

Ryan Downing (라이언 다우닝), chief information officer at Principal Financial Group, an investment company with $800 billion in assets under management, said, "High interest in Mythos is pushing all organisations to re-evaluate their security posture." He added, "In the past, there was always time between a vulnerability being found and being exploited, so many processes were built on that premise, but that is no longer the case."