Is Anthropic's new AI tool too dangerous?

Published: 2026-04-20 15:53:18 ET CSCO

Published on 04/20/2026 at 03:53 pm EDT

STORY: Tech firm Anthropic says its new AI model Mythos could be too dangerous to be released to the public.

Designed for defensive cybersecurity tasks, the company says the unreleased model can find and exploit serious software vulnerabilities at a level approaching and even, in some cases, exceeding human experts.

That's sparked fears that the same AI capabilities that could help defenders could also help attackers find new ways into critical systems.

:: What are the concerns around Mythos?

"This particular tool seems to be more powerful than anything else we've seen before and is able to identify vulnerabilities that others, including humans, have overlooked for decades."

Pia Hüsch is a research fellow at the Royal United Services Institute in England.

"So if the tool is as good as it seems, then maybe attackers could use it to layer several different layers of vulnerabilities or exploit several sectors in parallel, for example, power and water infrastructure, and then really launch an attack that is more powerful than anything we've seen before."

Cybersecurity involves a constant race between those trying to find and fix weaknesses and those trying to exploit them.

Mythos signals a significant step forward in that cyber arms race.

Industries already feeling vulnerable may now have more cause for alarm.

"I think the financial market has been particularly vocal about the fear of what this means to the banking sector, the potential to disrupt banking and processes in financial services. There's also a concern that this could be used by attackers to exploit vulnerabilities and critical infrastructure, which already relies on pretty outdated infrastructure in many cases already."

:: How has Anthropic reduced the potential risks?

To limit access, Anthropic placed the Mythos in a restricted initiative called Project Glasswing.

Project Glasswing brings together companies including Amazon Web Services, Apple, Cisco, CrowdStrike, Google, JPMorgan Chase, Microsoft, NVIDIA, and Palo Alto Networks, as well as more than 40 additional organizations that build or maintain critical software infrastructure.

Anthropic says that the purpose of Project Glasswing is to use Mythos to help secure critical software before others weaponize similar capabilities.

"The fact that this is one company that has the power to decide now which of its selected partners has access to this really illustrates how powerful these tech companies are. And at the moment it's really unclear who has access to this; and are there any European companies, for example, that have access to the tool. So I think there's many questions that are left unanswered at the moment that add to this hype and this perception of it being an absolute game changer."

:: What is next for cybersecurity?

The bigger issue may be less whether Mythos alone changes cybersecurity than whether it signals a broader shift.

Anthropic says it sees no reason to think Mythos represents a plateau, and says frontier capabilities are likely to advance substantially over the next few months.

CEO Dario Amodei in February suggested we may be on the cusp of a new era, where computer thinking surpasses the human intellect.

"We're increasingly close to what I've called a country of geniuses in a data center, a set of AI agents that are more capable than most humans at most things and can coordinate at superhuman speed."