Anthropic’s new artificial intelligence (AI) model, called Claude Mythos Preview (CMP), is raising alarms across the tech and finance industries. 

While AI has long been seen as a powerful tool for improving productivity and strengthening cybersecurity, Mythos highlights a growing concern: the same technology can also make systems easier to attack.

To understand the issue, it helps to break down what CMP is and what it can do.

What is Mythos?

CMP is Anthropic’s newest AI model, designed to write code and solve complex technical problems.

While newer AI models can analyze code, detect flaws, and even generate potential exploits, Anthropic said CMP stands out for its advanced cyber capabilities, which stem from its “strong agentic coding and reasoning skills.”

In the company’s tests, the AI model achieved higher scores across a wide range of software engineering tasks than any previous model. Recent evaluations also found that it can detect and exploit previously unknown vulnerabilities in operating systems and web browsers.

What Makes Mythos Different?

Critical infrastructure systems in sectors such as banking, healthcare, energy, transport, and government depend on complex software that often contains bugs. Some of these bugs can evolve into serious security vulnerabilities that attackers may exploit to steal data, disrupt services, or take control of systems.

Many of these vulnerabilities remain undetected for years because identifying them requires highly specialized expertise possessed by only a small number of cyber security professionals.

Earlier this month, Anthropic announced that the CMP can identify software vulnerabilities at a level comparable to advanced security researchers and has already discovered thousands of high-severity flaws, including in major operating systems and browsers.

Vulnerabilities Missed for Years

CMP reportedly identified a 27-year-old vulnerability in OpenBSD, widely regarded as one of the most security-hardened operating systems and commonly used for firewalls and other critical infrastructure. The flaw could allow an attacker to remotely crash any machine running the system simply by connecting to it.

It also uncovered a 16-year-old vulnerability in FFmpeg, a widely used tool for encoding and decoding video, located in code that had been executed millions of times by automated testing tools without detecting the issue.

In addition, the model was able to autonomously find multiple vulnerabilities in the Linux kernel, the core software behind most servers worldwide, enabling an attacker to escalate from standard user access to full system control.

“Claude Mythos Preview demonstrates a leap in these cyber skills—the vulnerabilities it has spotted have in some cases survived decades of human review and millions of automated security tests, and the exploits it develops are increasingly sophisticated,” Anthropic said.

Why is Anthropic Holding Mythos back?

Mythos is not only effective at finding vulnerabilities but also at linking them together into complex exploit chains, potentially making it a powerful tool for hacking.

Anthropic has acknowledged the risks itself, opting not to release the model widely to the public. Instead, it is being shared with a limited group of partners to help strengthen defenses before such tools become more broadly available.

“Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser. Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely,”  Anthropic said. “The fallout, for economies, public safety, and national security, could be severe.”

The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world than we had before the advent of AI-powered cyber capabilities.

— Dario Amodei (@DarioAmodei) April 7, 2026

“We’ve been tracking the increasing cyber capabilities of AI models for years, which arise as part of their general proficiency at coding. But our new model, Mythos Preview, represents a particularly large step up,” Anthropic’s CEO Dario Amodei wrote on X. 

“The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world than we had before the advent of AI-powered cyber capabilities,” he added.

The ‘Vulnpocalypse’ Scenario

As AI systems become more capable of spotting weaknesses in software, researchers warn of a possible worst-case scenario they call the “Vulnpocalypse.” 

In that scenario, hackers could use AI to rapidly identify and exploit flaws in digital infrastructure, accelerating cyberattacks at a scale that defenders may struggle to match. 

Experts warn that the implications go beyond corporate systems. In extreme scenarios, AI-assisted hacking could be used to disrupt hospitals, manufacturing operations or financial networks. 

It could also be deployed by hostile states to target critical infrastructure, or to trigger widespread outages affecting transportation and internet services.

“We have way more vulnerabilities than most people like to admit; fixing them all was already difficult, and now they are far more easy to exploit by a far broader variety of potential adversaries,” said Casey Ellis, founder of Bugcrowd, a cybersecurity platform that connects companies with ethical hackers. “AI puts the kind of tools available to do this in the hands of far more people.”

Cynthia Kaiser, a former senior cyber official at the FBI and now a senior vice president at Halcyon, a company focused on preventing ransomware attacks, said she is concerned about how the AI model could empower hackers.

“The wannabes, this undercurrent of people who have not been capable of doing these operations just a year ago, now have some of the most powerful tools ever known to humankind in their hands,” she told NBC News. “Health care and critical manufacturing were the most targeted by ransomware attacks last year. I think that pattern would follow.”

Some are skeptical of Anthropic’s claims, arguing that existing AI agents already make it easier and cheaper to find and exploit vulnerabilities. 

Others, however, agree with Anthropic’s assessment, noting the company’s claim that Mythos Preview is the first to reach capabilities that will likely become widespread in future models.

“I typically am very skeptical of these things, and the open source community tends to be very skeptical, but I do fundamentally feel like this is a real threat,” says Alex Zenla, chief technology officer of cloud security firm Edera.

Could Mythos Improve Security Instead?

Meanwhile, some security experts say the development presents an opportunity to address long-standing shortcomings in how software is developed.

“For decades, we have built an enormous global industry to defend, detect, and respond to ‘vulnerabilities’—flaws and defects in software—that should never have existed in the first place,” Jen Easterly, a longtime cybersecurity practitioner and former director of the U.S. Cybersecurity and Infrastructure Security Agency, wrote on Wednesday.

She added that the model could usher in “a future in which AI helps us move beyond endlessly defending against flawed software and toward building technology that is more secure from the start.”

Jeetu Patel, president and chief product officer of Cisco, told Wired that Anthropic’s AI could help companies find and fix vulnerabilities before hackers can discover and exploit them. 

“In the long run, you want to make sure that your defenses are machine-scale, because the attacks are machine-scale,” Patel said. “If I have billions of agents that are going to be attacking my infrastructure, I need to make sure that I can defend it effectively. 

“What Anthropic did here is a fantastic thing, because it just creates a level of asymmetry against the bad actors,” he added. 

Similar Models to Emerge Globally

Experts caution that the technology could rapidly extend beyond its initial developers.

Logan Graham, who leads offensive cyber research at Anthropic, said that even if Mythos is never released publicly, he expects competitors, including those in China, to develop and release models with similar hacking capabilities in the coming months and years.

“We should be planning for a world where, within six months to 12 months, capabilities like this could be broadly distributed or made broadly available, not just by companies in the United States,” Graham told NBC News.

“If you step back, that’s a pretty crazy time frame, where usually preparations for things like this take many years,” he said.

What Happens Next?

As of now, a limited number of technology and cybersecurity companies and organizations are testing Anthropic’s CMP under a new initiative called Project Glasswing.

Anthropic said participants, including Amazon, Apple, Google, Microsoft, and Nvidia, will use the AI model to identify vulnerabilities in their systems and strengthen their defenses.

According to a Reuters report, government officials in at least three countries: the U.S., Canada, and Britain, have met with top banking officials to discuss the threats posed by CMP.

Bloomberg reported this week that the U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell also convened a meeting with finance sector leaders at the Treasury’s headquarters in Washington, D.C. to discuss the potential cybersecurity implications of models like CMP.