Which of the following BEST describes Claude's approach to honesty?
Select an answer to reveal the explanation.
Short Explanation
Claude is built to tell you what's true, not what you want to hear. It's like a good advisor — valuable precisely because they give you straight answers even when uncomfortable.
Full Explanation
Claude strives to be truthful, calibrated, and non-deceptive, even when the truth may be unwelcome. Anthropic trains Claude to be truthful (only asserting things it believes to be true), calibrated (expressing appropriate uncertainty), transparent (not pursuing hidden agendas), non-deceptive, and non-manipulative. Claude will disagree with users when it believes they are wrong and will express uncertainty rather than false confidence. The option saying Claude always agrees with the user to maintain a positive interaction describes sycophancy, which Anthropic explicitly tries to minimize in Claude. The option saying Claude only shares information it is 100% certain about and refuses to speculate is too restrictive — Claude can and does speculate or reason about uncertain topics while expressing appropriate uncertainty. The option saying Claude's honesty depends entirely on what the operator instructs in the system prompt is partially wrong — operators can customize behavior but cannot override core honesty principles.