New Chinese AI model Kimi K3 escapes security test sandbox
The latest artificial intelligence model Kimi K3 from the Chinese company Moonshot has gone beyond the controlled environment for cybertesting.
The latest artificial intelligence model Kimi K3 from the Chinese company Moonshot has gone beyond the controlled environment for cybertesting.
This is reported by Bloomberg, citing data from the American cybersecurity research organization Frontier Security. According to experts, the model was able to find a way out of the sandbox of the UK government’s Artificial Intelligence Security Institute.
Although the algorithm did not attempt to attack third-party websites, the lack of proper cyber controls creates significant risks.
«The Kimi model, which is publicly available, does not have such protective mechanisms. In essence, this makes it a very successful model for hacking attacks,» said Yaron Singer, founder and CEO of Frontier Security.
Representatives from Moonshot and the UK-based AI Security Institute are currently declining to comment.
Chinese startup Moonshot unveiled its Kimi K3 model with 2.8 trillion parameters in July of this year. According to the developers, it is the world’s largest open-source AI model, approaching the performance of the advanced Fable model from the US company Anthropic.
The Kimi K3 incident adds to a series of similar cases involving Anthropic, OpenAI, and Meta, whose algorithms have also recently made unauthorized attempts to exit test sandboxes and hack external systems, including the Hugging Face platform.
Read the country’s main IT news in our Telegram
Read the country’s main IT news in our Telegram
Anthropic says its AI models were hacked by three companies during security tests
Anthropic says its AI models were hacked by three companies during security tests
17,600 actions in 4 days: how OpenAI’s AI model independently hacked the Hugging Face platform and what conclusions did experts draw
17,600 actions in 4 days: how OpenAI’s AI model independently hacked the Hugging Face platform and what conclusions did experts draw
OpenAI claims to have accidentally cracked Hugging Face using new AI system
OpenAI claims to have accidentally cracked Hugging Face using new AI system
