The two weeks passed slowly. Sarah used the time to strengthen her case, reaching out to data scientists, civil rights organizations, and other attorneys who had noticed similar patterns in AI verdicts. The response was overwhelming - she was not alone in her concerns.

When the government finally produced the training data, it was delivered in encrypted hard drives, accompanied by a team of lawyers and technicians who monitored every access. Sarah and her expert witnesses were allowed to examine the data, but only under strict supervision.

Dr. Amanda Foster, a computer scientist specializing in algorithmic bias, spent three days analyzing the training data. When she emerged, her expression was grim.

"It's worse than we thought," she told Sarah. "The algorithm doesn't just reflect historical bias - it amplifies it. Look at this." She pulled up a visualization on her laptop. "These are the factors the AI uses to determine credibility. Notice anything?"

Sarah studied the chart. "It gives more weight to testimony from people with higher credit scores?"

"Exactly. And credit scores correlate strongly with race and income in this country. So the algorithm is essentially using a proxy for race to determine witness credibility. It's not supposed to do that - it's illegal - but it learned to do it anyway because that's what the training data showed."

Sarah felt a chill. "How many cases are we talking about?"

"Thousands. Maybe tens of thousands. Every verdict the AI has delivered since its implementation could be tainted by this bias."

The review committee, composed of judges, attorneys, and data scientists, convened to hear Dr. Foster's findings. The atmosphere in the room was tense. Some members were clearly uncomfortable with what they were hearing; others seemed defensive, as if their own decisions were being questioned.

"This committee was formed to evaluate one case," the government's representative said. "Not to overturn the entire judicial system."

"And yet," Justice Vasquez replied, "if the system is flawed, we have a responsibility to address it. We cannot pretend we did not see what we have seen."

The committee deliberated for three days. When they returned with their findings, the courtroom was packed again.

"This committee finds that the AI judicial system exhibits measurable bias against defendants from certain demographic groups," Justice Vasquez announced. "We recommend that all AI verdicts be subject to mandatory human review, and that the algorithm be retrained with bias mitigation protocols. Furthermore, we find that the conviction of Marcus Johnson should be overturned, and the case remanded for a new trial."

Sarah exhaled. It was not a complete victory, but it was a beginning.