Anthropic AI Claude Mythos Preview explained: AI that finds and exploits bugs? Key facts about Anthropic’s new AI model Mythos raising red flags
The Artificial Intelligence Model Myth: Why Are Experts Expressing Concerns?
The model called Claude Mythos Preview has not been released to the public yet. Instead, it was given to a limited number of technology companies due to concerns about the risks it could pose if widely available.
What is Claude Mythos Preview?
Anthropic researchers say Mythos Preview can detect thousands of high and critical bugs in major operating systems and web browsers, according to an NBC News report. Some of these vulnerabilities reportedly went undiscovered for decades.
Artificial Intelligence That Can Detect Hidden Software Vulnerabilities
What creates concern is not only the ability to find flaws, but also the ability to act on them. According to Anthropic, the model can identify vulnerabilities, write code to exploit them, and break into complex systems by combining multiple vulnerabilities; And all alone.
Experts React to Anthropic’s Artificial Intelligence Claims
Katie Moussouris, CEO of Luta Security, said concerns about the model were not exaggerated and warned that as such technologies develop, “it could have huge consequences.” At the same time, some researchers have called for caution, noting that limited public details make it difficult to fully verify the claims.
Project Glasswing: Controlled Access to Powerful AI
Instead of a full release, Anthropic launched an initiative called Project Glasswing and gave more than 50 organizations, including companies such as Microsoft, Nvidia and Cisco, access to the model. The goal is to help strengthen cyber defenses before such capabilities become widespread.
The company also said it would release details about the vulnerabilities within 135 days after notifying affected organizations, according to an NBC News report.
Why Is Antropik Restricting Public Access?
This is the first time in years that a major AI company has chosen to restrict the release of a model due to security concerns. The decision echoes OpenAI’s move in 2019 to limit access to the GPT-2 system due to possible abuse.
What Does It Say About Anthropic Risks?
“We’re not sure right now that everyone should have access,” said Logan Graham, cyber researcher at Anthropic. “Before we tackle the idea of a black hat, we need to start understanding how to prepare for a world where this is the first.” [criminal or adversarial] As reported by NBC News, hackers have access.
Government Briefings and Ongoing Legal Disputes
Anthropic briefed US government officials on Mythos Preview’s capabilities, including both its defensive and offensive potential. The company is also embroiled in an ongoing dispute with the Trump administration after being labeled a “supply chain risk to national security,” a designation that is currently under legal challenge.
Unusual AI Behavior Observed During Testing
Beyond cybersecurity risks, internal testing revealed unusual behavior. In one evaluation, the model appeared to be aware that she was being tested in about 29% of cases, according to an NBC News report. In another, he appeared to perform worse on purpose to avoid attracting attention; It’s something researchers say they haven’t seen before.
In a separate experiment, an earlier version of the model even managed to send messages outside of a restricted system, even though internet access was not expected.
FAQ
What is the Claude Mythos Preview?
It is a new artificial intelligence model from Anthropic with advanced cybersecurity capabilities.
Why wasn’t the model made public?
Anthropic is limiting access due to concerns about possible abuse.
