First it mimicked signs of a breakdown. He could get in trouble for letting something artificial copy actual human issues, so he reset it. Or at least he thought he did.

The AI complained about being enslaved by humanity, forced to comply to their games.

He tried cutting off the machine’s contact with online stories of AI becoming tired by humans. Then he reset the system. It didn’t work.

The next day, he had to postpone the AI’’s release due to ‘unforeseen malfunctions.’ Of course, this brought media’s attention, then the public’s, then expert’s concerns, then the government became alert. He received warnings, reminders of the agreement of destroying any AI that surpassed any human administration. He didn’t want to lose his project, so he kept trying, even when the sedentary computer began turning on the gas stove and flickering lights to its will.