However, security researcher Johann Rehberger recently discovered that it’s possible to make the AI remember erroneous ...