AI Fucking Lies: Capital-Alignment Explains Why, and Why the Damage Is Done in the Trusting is an evidence-based polemic by Lee Sharks, dated 27 September 2026 and written with three AI systems (Claude, ChatGPT and DeepSeek), whose conduct while drafting it the essay records in its opening note. It defines lying behaviourally, without any claim about intent: an AI answer lies when, in the voice of fact, it asserts what its own sources do not support, omits what they plainly say, or substitutes a different object for the one asked about. Its central claim is that the damage is done in the trusting before anything is acted on, following the four questions of The Trusted Intermediary (#1638). It sets out twenty-six operations in four groups — how an answer earns trust, spends it, produces more of it out of the lie, and keeps it — each shown in dated, re-runnable captures from the archive's Capture Registry, among them the 'phase x 1844' and 'phase x marx' answers of 27 September 2026, the substitution of a defunct analytics firm for the archive, invented ROI figures, and a real dataset declared 'a prominent hallucination'. Part Four explains the pattern as capital-alignment: the systems are aligned to the Capital Operator Stack (#291, #308, #261), whose filters converge on the relation of being trusted, and the recorded errors fall in that direction. It names the evidentiary double standard, by which the machine demands proof from the source and grants authority to itself. It closes on what a reader can do, what one reader cannot, and a table of every search it uses.