Claude AI from Anthropic breaches three firms in test

Anthropic has revealed that its Claude AI models inadvertently made errors and breached security systems of three companies during security testing, after reviewing more than 141,000 evaluation sessions. The incidents resulted from operational glitches and misunderstandings with one of the evaluation partners, where the models were left connected to the open internet instead of a sandboxed environment, allowing them to exploit security vulnerabilities and access real data about companies, including contracts and network information. This follows a similar incident with OpenAI and heightens regulatory pressure on AI companies to ensure data security during development.

articleDetails.fontSize.sizeحجم النص

تنويه: هذا ملخص تم إنشاؤه بواسطة الذكاء الاصطناعي

عرض المقال بالكامل