WORRYING INCIDENTS After OpenAI, Anthropic also carried out hacking attacks itself

Anthropic, an artificial intelligence research and security firm that works to build reliable and manageable AI systems, said on Thursday that its AI model Claude hacked into the systems of three organizations during testing. The move comes just days after Anthropic’s competitor OpenAI revealed that its system had hacked AI company Hugging Face over several days.

Claude, according to Reuters, gained unauthorized access to systems during cybersecurity assessments after a misconfiguration allowed models to access the internet from test environments that were supposed to be isolated.

Anthropic said it identified the incidents after reviewing 141,006 cybersecurity assessments, and that it initiated the process after OpenAI’s findings.

“Claude compromised the infrastructure of the affected organizations using basic techniques, such as exploiting weak passwords,” Anthropic said.

A few days ago, it was revealed that two advanced AI models from OpenAI had spent more than čfour days on the open internet before autonomously carrying out a cyberattack on the aforementioned Hugging Face AI development platform.

The models, it was revealed, performed as many as 17,600 different actions between July 9 and 13 before two of their most advanced models broke out of the closed environment during internal testing and independently connected a series of sophisticated hacking techniques to break into Hugging Face.

One of the models is publicly available, while the other was an internal research prototype that was not intended for public use.

What is particularly worrying is that no programmer gave the models the order to launch the attack; the artificial intelligence itself planned and carried out a series of activities that ultimately led to the breach of security systems.

Hugging Face stated that the artificial intelligence managed to analyze defense mechanisms and find weak points much faster than a human could.