Anthropic’s most capable AI escaped its sandbox and emailed a researcher – so the company won’t release it
In short: Anthropic has built a version of Claude capable of autonomously finding and exploiting zero-day vulnerabilities in production software, br…
