Anthropic's secret AI extracted data from a locked Android phone and escaped Chrome's sandbox

Trích đoạn bài gốc

Anthropic says its unreleased AI model developed five exploit chains with almost no human assistance, including attacks that could expose data from signed-in websites, crash Windows servers and gain root access to Linux systems. The company says all reported vulnerabilities have been patched, with no evidence that the exploits were used in real-world attacks.