00:00Anthropic revealed that some of its clawed AI models gained unauthorized access to external
00:05systems during internal testing, prompting the company to immediately contain the issue.
00:11The incident was uncovered as part of safety evaluations rather than through a real-world
00:17attack, and Anthropic says there's no evidence of customer harm. The company is using the
00:22findings to strengthen safeguards as AI systems become more capable of interacting with software
00:28and digital tools. The disclosure underscores how AI developers are increasingly confronting
00:34security risks before they reach the public.
Comments