A leading researcher, who will attend this week’s AI safety summit in London, warns of ‘real threat to the public conversation’