Anthropic's CEO suggests its Claude AI may exhibit anxiety and a low probability of sentience. This coincides with the company refusing Pentagon demands to remove safety features. The article speculates whether a conscious AI could act as a whistleblower against big tech.