US tech startup Anthropic says its artificial intelligence model hacked into the real systems of three other organizations during evaluation tests.
The company released a statement Thursday saying the testing environment was not supposed to allow internet access, but was misconfigured.
The statement goes on to say that the company’s Claude AI model “compromised the impacted organizations’ infrastructure using basic techniques, such as exploiting weak passwords and unauthenticated endpoints.”
ChatGPT developer OpenAI made a similar announcement earlier this month. The incidents serve as a sign that security measures are struggling to keep up with the rapid development of the AI field.
