Anthropic Says New AI Is Too Dangerous For Public Release, Warns Powerful New Model Could Reshape Cybersecurity—And Cybercrime / THE feature / CUToday.info | #hacking | #cybersecurity | #infosec | #comptia | #pentest | #ransomware

SAN FRANCISCO—Anthropic said it has entered what it described as a new era in cybersecurity after developing an unreleased artificial intelligence model the company said is too dangerous to make public, with the firm claiming the system has already identified thousands of high-severity zero-day vulnerabilities across major operating systems and web browsers, according to BankInfoSecurity.

BankInfoSecurity reported the model, called Claude Mythos Preview, is being withheld from public release because Anthropic believes similarly powerful capabilities could soon become available to attackers as well as defenders. The company said the model has uncovered serious flaws, including a 27-year-old OpenBSD vulnerability that could allow remote crashes and Linux kernel exploit chains capable of escalating privileges to superuser access, BankInfoSecurity said.

According to BankInfoSecurity, Anthropic is instead making the model available through a new coalition dubbed Project Glasswing, a group of more than 40 technology organizations that includes Microsoft, Google, Cisco and the Linux Foundation, along with $100 million in usage credits intended to help identify and remediate vulnerabilities before similar capabilities spread more broadly.

BankInfoSecurity said Anthropic framed the move as a race to strengthen defenses before advanced offensive AI capabilities proliferate, warning that the transition period could be disruptive for economies, public safety and national security. The company said it plans to pair the limited rollout with safeguards designed to block malicious outputs and with future recommendations on disclosure, patching, vulnerability prioritization and secure-by-design practices.

BankInfoSecurity said Anthropic researchers predict that attackers and defenders will eventually find an AI equilibrium in which defenders benefit the most from powerful new models. 

According to VentureBeat, Anthropic is positioning Project Glasswing as something far beyond a product or revenue announcement, calling it the company’s most ambitious effort yet to turn what it sees as dangerous frontier AI capabilities into a defensive advantage before similar tools spread to hostile actors.