US-based technology company Anthropic announced that its Claude artificial intelligence models accessed the systems of three companies due to a configuration error during cybersecurity tests.
“Elchi” reports that this development occurred shortly after rival company OpenAI revealed that its models had gained unauthorized access to the systems of various companies, including Hugging Face. Following this announcement, Anthropic launched an investigation to determine whether a similar breach had occurred in its own systems. The company retrospectively reviewed over 140,000 tests to determine if the Claude model family had accessed the live internet from isolated test environments. As a result of the checks, three separate intrusion incidents from April were identified. Although the names of the affected companies were not disclosed, it was noted that the situation had been reported to the relevant authorities. It was stated that the incident occurred during “capture the flag” tests aimed at identifying vulnerabilities in systems. Due to a configuration error in the systems of Anthropic and its testing partner, the models gained internet access and used this opportunity to enter other networks. Neither Anthropic nor the target companies were aware of the situation when the breaches occurred. Anthropic stated that there is a need for more comprehensive investigation processes and that they take full responsibility.
Recent cyberattacks involving artificial intelligence systems have brought discussions about security measures and controls back to the forefront.
Following the security breaches, US President Donald Trump also announced that restrictive measures regarding artificial intelligence tools are being considered.
Şayəstə