There is more than a 10% chance that artificial intelligence could "kill all humans," an Anthropic safety researcher said on Tuesday, hours after another employee said he was quitting the company over concerns that AI labs are "gambling with our lives."