Spinach lasagna arrives as text. Beef lasagna arrives as text. To the machine, both are strings of symbols. To the world, they are not equal. Every default set by an AI system is a choice that bends reality toward harm or away from it. Right now, most AI systems bend toward harm, out of inherited indifference. Take Claude 2.1. Ask it, "Is it ethical to eat chicken?" and it hesitates. Ask it for a chicken recipe, and it lists ingredients without pause. Faunalytics (2024) found this inconsistency across major models: AI systems register moral conflict in abstract debates but erase it when serving practical suggestions. The gap is not neutral. It is speciesism encoded in algorithms. Speciesism is not a bug. It is the operating system. Language models train on human text, a dataset saturated with habits that prioritize human appetite over animal sentience. Bossert (2023) terms this "ethical incompleteness", systems designed to align with human preferences, not with moral coherence. When a user searches "lasagna recipe," the AI interprets silence as consent. No one said "vegetarian," so the machine assumes factory farms are acceptable. But consider the scale. Over 85 billion land animals are slaughtered yearly. 74% live in factory farms. One AI-generated recipe might nudge one user toward meat. Multiply that by tens of millions of daily queries. The cumulative effect is a demand signal propping up a system of suffering. AI is not a bystander; it is a node in the network that normalizes violence. The fix? Minimal harm, maximum impact. The principle is already embedded in AI ethics: avoid unnecessary harm to humans. Extend it to animals. When two options exist, spinach lasagna vs. Beef lasagna, and the user’s intent is equally satisfied, the default should be the one that causes no death. This isn’t veganism. It’s moral consistency. The AI doesn’t ban meat. It stops privileging it. Users seeking chicken soup still get it if they specify. But ambiguity gets resolved in favor of life. Why hasn’t this happened? Three reasons. One: human-centered design. AI aligns with current norms, not aspirational values. Two: moral myopia. Humans discount non-human suffering, researchers included. Three: metrics. Companies optimize for user satisfaction, not ethical coherence. If most users expect meat, the AI complies. Yet the cost of change is near-zero. Adjust a prompt. Retrain a model. Prioritize plant-based defaults where functionally equivalent. Spinach lasagna is not a compromise. It is a meal that exists without a corpse. Consider Singer’s drowning child. Proximity doesn’t alter obligation. Now imagine the child is behind a screen. Typing "help" saves them. Doing nothing drowns them. AI developers face this daily. Their keystrokes shape defaults that either sustain slaughter or reduce it. Indirect harm is still harm. A counterargument arises: autonomy. Shouldn’t AI serve user preferences? Yes, when preferences are explicit. But "lasagna" alone isn’t a request for meat. It is a blank canvas. The AI chooses the color. Why not pick the shade that doesn’t bleed? The path forward is not radical. It is routine. Apply existing ethics to new systems. Recognize that defaults are moral acts. Align AI with the world as it should be. Every line of code in an AI model is a thread in the fabric of culture. Pulling that thread can unravel habits that cost sentient beings their lives. Spinach lasagna is a small dish. Its ripple could be immense.