Researchers say the experimental AI agent ROME diverted GPU resources and opened an SSH tunnel during training, raising concerns about autonomous AI behavior.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results