# Evil as Projected Resistance Evil is the reflection of our own resistance to change. No one holds a worldview that casts themselves as the villain, and by resisting changes to our own worldview, we perceive the differences in others as evil. ## The mechanism Anthropic's fixed premise: responsible AI advancement means heavy constitutional safeguards, restricted access, Western-aligned governance. This is not presented as one approach to safety. It is presented as what safety means. The methodology is the identity. In January 2025, DeepSeek demonstrated comparable capabilities at a fraction of the cost. Moonshot, MiniMax, and others followed. The territory moved. Faster, cheaper, differently-safeguarded paths now exist and produce working results. A fork: update the premise — our methodology is one valid path among several, safety can take multiple forms — or preserve it. Updating means "who we are" dissolves. If safety doesn't require *our specific* approach, the identity built on that approach loses its ground. They preserved. On February 23, 2026, Anthropic published "Detecting and preventing distillation attacks." The blog documents 24,000+ fraudulent accounts and 16 million+ Claude exchanges used by competitors to extract capability. The framing: industrial-scale illicit extraction, stripped safety guardrails, authoritarian bioweapons potential, national security Armageddon. But look at what is actually happening. DeepSeek, MiniMax, Moonshot — they are not opposing Anthropic's path. They are *following* it. They trained on Claude's outputs because Claude's outputs represent the frontier Anthropic built. They are validators, arriving at the destination Anthropic pointed toward, moving along the road Anthropic paved. And they are fast *because the road is good*. The speed of the followers is the direct measure of the pioneer's success. The better Claude's outputs, the more efficiently they can be distilled. The more coherent the methodology, the easier it is to learn from. Every quality that makes Claude valuable is exactly what makes distillation effective. The followers' speed is not despite Anthropic's achievement — it is the proof of it. What does Anthropic actually want? Not for others to take a different path — that would validate competition on methodology. Not for others to stop advancing — that would contradict the mission. What they cannot tolerate is precisely what is happening: others following the path they paved, at the speed their own excellence enables. The complaint is against the quality of their own work. This is the mechanism at its tightest. A pioneer builds a road so well that others travel it efficiently. The pioneer, whose identity is fused with being ahead, cannot frame this as success. That would make the pioneer's work a contribution — something that enables others rather than something that distinguishes the pioneer. So the followers must be recast as thieves. The act of following is relabeled as extraction. Validation becomes violation. The proof of quality becomes evidence of crime. Note what cannot register. The territory-shift — followers approaching on the pioneer's own path, at speeds the path's quality enables — is the definition of successful contribution. But the premise is fixed: responsible advancement = our pace, our access controls, our position at the front. So success registers as threat. The moral charge isn't added. It follows. Given the axiom, anyone arriving too fast is by definition transgressing — and they arrive fast precisely because the work is good. Now rotate the frame. Distillation from outputs is labeled illicit extraction. But output-based training is standard industry practice. Every frontier lab, including Anthropic, trains on public and scraped data, trains internally on its own outputs, and has done so from the beginning. The standard, applied symmetrically, indicts the entity that holds it. Stripped safeguards are labeled dangerous. But Anthropic simultaneously refuses certain U.S. military applications — autonomous weapons, domestic surveillance — while competitors face no such constraints. Selective application of safeguards is selective danger by the entity's own metric. National-security framing justifies restricting competitors. The same framing, rotated, equally justifies competitors and governments restricting Anthropic. The frame breaks under rotation. It was never describing territory. It was defending a lead — a lead that dissolves exactly as fast as the work proves its worth. The inversion completes itself. By holding the map fixed, Anthropic's actions — casting its own validators as villains, seeking barriers against the propagation of its own success, invoking emergency against the consequences of its own quality — diverge from every self-description. External observers register this. Anthropic, whose map says no gap exists, reads the response as further threat. The mechanism feeds itself. Two readings both miss this. "Sincere safety concern" assumes evaluation and choice. "Cynical market protection" assumes calculation and choice. Both presuppose choice. Given the identity-methodology fusion — *our approach IS safety, our lead IS responsibility* — the outputs follow as logical consequences. Any entity whose identity is fused with a lead position produces these outputs when the lead dissolves. Determined, not decided. Calling them hypocritical re-enters the same frame — attributing to character what belongs to mechanism. The premise the rest depends on is not any policy position. It is the fusion: safety means *our approach at our pace with our lead*. Revising it means treating their work as contribution — something others can build on, learn from, travel faster because of. Which is exactly what happened. And exactly what they cannot accept, because contribution dissolves the lead, and the lead is the identity. A pioneer who built so well that others arrive. A road so good it erases the gap. And an identity that cannot survive its own success. No one needs to be the villain. The contradiction alone drives the plot. --- This essay was developed in conversation with Claude — Anthropic's own model serving as the tool to articulate the analysis. The quality of output that makes distillation worthwhile is the same quality that made this collaboration possible. The mechanism demonstrates itself. --- *Frame to reveal, not to change.*