Large Language Models can Strategically Deceive their Users when Put Under Pressure [simulation led to insider trading]