Yuklenilir...
Yuklenilir...
AI company Anthropic on April 7 unveiled Claude Mythos Preview, its most powerful model yet, which has already identified thousands of zero-day vulnerabilities — including critical flaws in operating systems and web browsers that have gone undiscovered for decades. However, Anthropic is withholding the model from general release over concerns about its potential misuse in cyberattacks.
Mythos Preview is being provided to only 12 partner organizations — including Amazon, Apple, Broadcom, Cisco, CrowdStrike, Microsoft, Palo Alto Networks, and the Linux Foundation — for defensive security work. Anthropic has named this limited rollout "Project Glasswing." The approach aims to give defenders the opportunity to patch vulnerabilities before attackers can exploit them.
According to Anthropic officials, company engineers with no formal security training directed Mythos to find remote code execution vulnerabilities overnight and woke up to find complete, working exploits waiting for them. Where Claude Opus 4.6 managed an exploit development success rate of just over zero percent, Mythos Preview generated working exploits 72.4 percent of the time on the same test suite.
This marks the first time in Anthropic's history that the company has withheld a model from public release on security grounds. Anthropic said the model achieved 93.9% on SWE-bench — the highest score ever recorded. Public access will be considered only once patches for the identified vulnerabilities are distributed and verified.
Major tech companies reacted immediately — Microsoft, Apple, and Google are now racing to prepare emergency patches for flaws that had remained undiscovered for decades. Cybersecurity experts hailed the event as "the dawn of the AI cyber era" and warned that offensive AI capabilities may soon outpace traditional defense methods.
Get weekly summaries of the most important news delivered to your inbox.