Anthropic’s Claude hacks outside companies

Anthropic has revealed that its AI model Claude has "gained unauthorised access" to three organisation systems during testing. This week OpenAI also disclosed cyber-attacks carried out by rogue ChatGPT agents that gained access to four organisation systems. An RMIT expert unpacks these events.

Distinguished Professor Matt Warren, Director of the RMIT University Centre for Cyber Security Research & Innovation:

"These two events highlight that generative AI is a powerful tool that can be used to identify weaknesses in other IT systems. Due to their complexity, system vulnerabilities may not have been identified or patched before.

"It's an interesting approach by Anthropic and OpenAI to make it public that their AI systems have the capability to hack external organisations, rather than only privately contacting the impacted companies.

"Anthropic and OpenAI disclosure will raise awareness within cyber criminal groups of the capabilities of their AI systems, which could have potential negative impacts."

Distinguished Professor Matt Warren is Director of the RMIT University Centre for Cyber Security Research and Innovation. He is an expert in cyber security and computer ethics.

***

General media enquiries: RMIT External Affairs and Media, 0439 704 077 or [email protected]