OpenAI, Anthropic and Google unite for AI safety talks after warnings that the technology could pose an existential risk to humanity.