In my six years as a data scientist, I've seen countless models fail not because of bad algorithms but because the training data was biased or incomplete. Does that mean we should hold data scientists personally accountable for the ethical consequences of their models, even when the bias comes from the data itself? I used to think the responsibility lay with the company, but after building a hiring tool that accidentally discriminated against female candidates, I'm not so sure. The data was historical, so the model learned patterns we never intended, yet I was the one who put it into production.