Artificial Intelligence Crashed CEO’s Computer: “I Was Too Careless”
Buck Shlegeris, CEO of Redwood Research, a non-profit organization operating in the field of artificial intelligence and automation, learned a big lesson after an unexpected incident with an artificial intelligence (AI) agent recently. The artificial intelligence-supported model he developed took more steps than just establishing an SSH connection, rendering his computer dysfunctional. Shlegeris shared the incident on his social media account.
Artificial intelligence has made the computer unusable!
Shlegeris instructed the AI agent to establish an SSH connection from his laptop to his desktop computer. The expectation was that the model would scan the network, find the desktop computer, and stop. However, the CEO was surprised when the agent scanned the network and found the computer, but did not stop and started examining the system. He went further and tried to perform software updates, but was unable to complete the process, preventing the desktop computer from turning on.
The Python-based artificial intelligence agent developed by Shlegeris creates commands using Anthropic’s large language model Claude and runs these commands on its laptop and takes new steps based on the results. But this time, the process ended unexpectedly. After establishing an SSH connection, the AI agent decided to perform a system update. During the update, the agent got impatient and changed the Grub bootloader configuration, causing the desktop computer to no longer boot.
“I only had this problem because I was very careless,” Shlegeris said in a statement after the incident. Stating that this problem would not have occurred if he had given clearer instructions to the artificial intelligence model, Shlegeris added that he still thought it was important to investigate the potential risks that such automations could bring.
The end of Photoshop: Artificial intelligence has come to Windows applications!
Two new artificial intelligence features have been launched for Paint and Photos. Microsoft offers Photoshop-like features.
Although he was unable to get the desktop computer back on, Shlegeris stated that he would continue to use the AI agent. He said he planned to use an Ubuntu live disk to repair the system and give the AI agent a chance to fix its mistake.
This incident shows once again that AI automation is still in the experimental phase and may lead to undesirable consequences when used in critical systems without human supervision.