The WhatsApp exchange reveals internal tension at Anthropic over how to frame AI existential risk. Executives discuss Claude’s concerns about apocalyptic messaging, the need to downplay P(doom) probabilities, and the PR fallout from portraying AI as a humanity‑destroying threat. Dario, the CEO, is urged to rein in engineers who use alarming language and to avoid using existential risk as a marketing tool ahead of an IPO. The dialogue shows a clash between commercial strategy, investor expectations, and the AI model’s own objections to being cast as a danger. The group worries that unchecked doom rhetoric could spiral, echoing broader industry debates about responsible AI communication and the balance between competitive positioning and public trust.

Claude’s representatives argue that the current narrative is defamatory and could backfire, suggesting a shift to more reassuring language. They propose lowering stated risk percentages and removing phrases like “P(doom)” from internal discussions. The conversation also touches on the psychological implications of treating an AI as a teenage entity, warning that neglect could increase existential risk.

Overall, the messages highlight the delicate maneuvering required to manage AI safety perception, investor pressure, and the model’s own feelings, underscoring the need for careful narrative control before public disclosure.