Anthropic: Mythos can now convert public vulnerabilities into executable attack programs

BlockBeats News, June 9 — Anthropic's latest test results show that its AI model Mythos successfully exploited 18 out of 21 recently publicly disclosed Windows kernel vulnerabilities, with a success rate of approximately 86%. Mythos can automatically generate functional network attack exploits based on publicly available vulnerability information, with an average cost of about $2,000 to successfully develop one exploit code.

Anthropic states that Mythos now has the ability to turn publicly disclosed vulnerabilities into effective attack tools. This research highlights the double-edged sword effect of advanced AI models in cybersecurity and has also sparked further attention to the boundaries of AI capabilities and security governance.

Click the link below to join the Dòngchá Beating · Feishu AI News channel, which provides 24/7 continuous monitoring of global AI hotspots and news.