Anthropic Acknowledges Data Risks in New Claude Feature

Most generative AI services can work with personal or work-related data and files to some extent. This can save time and effort but poses a risk of sensitive information being shared with unauthorized parties.

A recent example is Anthropic's Claude AI. On Tuesday, the company announced that Claude AI can now create and edit Word documents, Excel spreadsheets, PowerPoint slides, and PDFs. This feature is currently available for Claude Max, Team, and Enterprise subscribers, with plans to extend it to Pro users soon.

However, there are risks involved in such interactions. Anthropic acknowledged that the feature gives Claude internet access, which may put data at risk. While the feature operates in a sandboxed environment with limited internet access, attackers could exploit this to run malicious code or access sensitive data.

Users are advised to monitor Claude when using the file creation feature. If unexpected data usage or access is detected, it should be stopped immediately, and issues should be reported. This reflects a common challenge in the generative AI industry, where providers have been slow to address such threats.

Anthropic has outlined several security measures for Claude users. Continuous security testing and red-teaming are in place, and organizations are encouraged to evaluate these protections against their specific security needs before enabling the feature.