Anthropic CEO and co-founder Dario Amodei speaks at the 56th World Economic Forum Annual Meeting in Davos, Switzerland, January 20, 2026.
Dennis Bariboos | Reuters
Anthropic on Tuesday announced an advanced artificial intelligence model that will be rolled out to a select group of companies as part of a new cybersecurity initiative called Project Glasswing.
The model, called Claude Mythos Preview, is excellent at identifying weaknesses and security flaws within software, and Anthropic has restricted access to prevent malicious actors from exploiting its features, the company said.
the human said apple, google, microsoft, Nvidia and Amazon Web Services is one of the initial launch partners of the project and will be able to use the model for defensive security work. It also includes more than 40 other companies. cloud strike and palo alto networksis also participating, Antropic said.
“There was a lot of internal consideration,” Diane Penn, director of research product management at Anthropic, told CNBC in an interview. “We see this as a first step to give many cyber defenders a head start on a subject that will become increasingly important in the future.”
Anthropic’s announcement comes after Fortune discovered a description of the model in a publicly accessible data cache late last month. Cybersecurity stocks fell after reports said the model was equipped with advanced cyber features that also pose serious risks.
The iShares Cybersecurity ETF was nearly flat in intraday trading on Tuesday.
“Getting this wrong is clearly dangerous, but getting it right creates a real opportunity to build an Internet and world that is fundamentally more secure than before the advent of AI-powered cyber capabilities,” CEO Dario Amodei wrote in a post on X promoting the Project Glasswing rollout.
Anthropic was founded in 2021 by a group of researchers and executives who defected from OpenAI over concerns about its direction and stance on safety.
The company has spent years carefully building a reputation as a company dedicated to responsible AI adoption, announcing Project Glasswing just weeks after a high-profile clash with the Pentagon over safety spilled into the public domain.
Anthropic said it is in “ongoing discussions” with U.S. government officials about Claude Mythos Preview’s cyber capabilities.
Anthropic employees decided on the name Project Glasswing, a metaphor for a transparent butterfly to a “relatively invisible” software vulnerability, Penn said.
According to Anthropic, Claude Mythos Preview can detect bugs that were previously difficult to detect, including some serious bugs. In one case, the company said, the model identified a 27-year-old bug in OpenBSD. According to the company’s website, OpenBSD is a security-focused operating system.
Anthropic said Claude Mythos Preview is a general-purpose model that is not specifically trained for cybersecurity, and its improved cyber capabilities are the result of strong coding and reasoning skills. The company said it has no plans to make the model publicly available, but the goal is to eventually learn how to deploy Mythos class models at scale.
The companies participating in Project Glasswing all build or maintain critical software infrastructure, and Anthropic said it seeks to use its model to secure both first-party and open-source systems. Anthropic has committed up to $100 million in usage credits for these efforts, but said partners will pay a fee if they use the model beyond that threshold.
Newton Cheng, Frontier Red Team cyber lead at Anthropic, said the company wants businesses to get used to taking advantage of these features before they become widely available. The company said it strives to avoid “recklessly or irresponsibly” releasing models that could be exploited by adversaries.
“Cybersecurity is an area where this broad capability improvement comes with potential risks, so we need to closely monitor what’s happening there,” Chen said in an interview.
WATCH: Big Tech’s Alex Kantrowitz says he thinks cyber stock panic is overdone
