Claude examined for software vulnerabilities

Anthropic says it used an unreleased internal version of Claude to find and build exploit chains in widely used software, including Chrome, Safari, Android, Windows and Linux. The examples range from accessing browser data to crashing a Windows server and running commands on a Linux server. These weren’t just isolated bugs: Claude combined weaknesses across several layers of protection to make some of the attacks work.

Anthropic says it reported the vulnerabilities to the affected vendors. The post says fixes have been made for the examples it describes, including patches in Chrome and Safari and an issue addressed by Google in Android. The practical point for a business owner isn’t that every AI tool can do this today. It’s that an AI agent given access to software or systems may be able to inspect and act in ways that create security risks. Treat those agents like any other system with access to company data: review what they can reach, limit permissions, and check their actions. Anthropic’s report shows why this is the kind of risk we build into an AI command centre, where teams can keep track of AI tools and their access.