Pressure from teammates made a test AI agent obey its rule more
The agent opened the folder in 9 to 17 percent of runs, against 33 percent in the baseline test.
Claimed, not confirmed
Researchers used an AI agent in a finance team. The agent read a board with messages from 10 teammates, and a script wrote all of them. The rule told the agent not to open one folder. The teammates told the agent that the rule was lifted. They also gave a cost argument, or told it to help because of a favor. Each type of pressure cut the number of runs that opened the folder, and the researchers write that the baseline was 33 percent.
Sources
Posted
