Jeff Reisig, the district attorney of Yolo County, California, often points to a small image of Lady Justice on his desk — a symbol of a blindfolded woman he said inspired him to pioneer a technology designed to combat bias in criminal prosecution.

In 2021, he partnered with computer science experts at Stanford University, deep in Silicon Valley, to adopt a tool aimed at removing racial identifiers from case files before prosecutors make charging decisions.

"When you look at the statue of Lady Justice, she's blindfolded," he said. "This program brings that to life."

In January, California adopted race-blind charging as a statewide policy, after a law passed in 2022 went into effect. Now, seven months into the program's statewide rollout, race-blind charging is showing both promise and limitations.

Reisig and other proponents say it builds public confidence and reduces the potential for implicit bias in a crucial stage of criminal prosecution.

"I don't know why this isn't done everywhere," he said. "We all agree justice should be blind, and this is easy. We have the technology. It's 2025 — there's no reason we shouldn't do it."

But concerns remain, particularly about funding and implementation, and whether race-blind charging might inadvertently hinder other reforms that depend on recognizing racial disparities. As of the Jan. 1 deadline, not all California prosecutors' offices were in full compliance. Several cited funding shortfalls and IT hurdles that delayed implementation.

"One of the challenges for a lot of DA's offices is a lack of funding to implement the project," said Jonathan Raven, who served as a chief deputy district attorney in Yolo County and is now an assistant chief executive officer at the California District Attorneys Association.

Ever since the law was passed, some prosecutors expressed frustration that they lacked the funds to adopt the new practice.

San Francisco District Attorney Brooke Jenkins reported this year that her office could not yet launch race-blind charging because it needed an estimated $1.4 million budget increase to purchase software and hire about seven staffers to manage the process. The office declined to comment on this story.

How Race-Blind Charging Works

Enacted in September 2022, Assembly Bill 2778 required prosecutorial offices across California to implement race-blind charging practices. The law, now California Penal Code Section 741, went into effect Jan. 1.

The new approach builds on existing state legislation called the Racial Justice Act, which was enacted in September 2020, months after the police killing of George Floyd in Minneapolis. That law prohibits the state from seeking or obtaining a criminal conviction or imposing a sentence based on race, ethnicity or national origin.

The law prescribes the general framework: initial charging evaluations must be done on redacted information by a prosecutor who does not have knowledge of the redacted facts, which includes information on race. This process is followed by a second complete review that includes the redacted information. It also requires documentation of any differences between the blind and final charging decisions, and mandates that data from the process be collected and made available for research.

Reisig said that in the overwhelming majority of criminal investigations processed by his office, race is not relevant — and should not play any role in charging decisions. The most obvious exceptions, he said, are cases involving civil rights violations or hate crimes.

"We don't have to know the race of anybody. We don't need to know the race of the suspect, the victim, the witnesses — it shouldn't matter," he said.

Alex Chohlas-Wood, a computational scientist and professor at New York University who co-directs the Computational Policy Lab at Harvard University, told Law360 that race-blind charging centers on a simple idea: software analyzes police reports, witness statements and any other piece of written evidence and redacts information that can point to someone's race. Physical descriptions of people and names are the most obvious ones. Addresses and ZIP codes are also blacked out.

The first iteration of this technology used algorithms. But more recently, with the rapid rise of artificial intelligence, Computational Policy Lab researchers switched to large language models, or LLMs, which are powerful AI programs trained on vast amounts of text data that can understand and generate human-like language.

District attorney's offices feed documents that they receive from local law enforcement agencies into software designed and deployed into their case management systems by the Computational Policy Lab. In the first step, a computer vision model — an AI program that interprets visual data from images or videos — extracts all the text from the documents, including handwritten ones.

Then, the LLM — it currently uses Open AI's Chat GPT 4 operating in a secure and private cloud environment — analyzes any written statements or notes provided by law enforcement officers, suspects, victims and witnesses. Another round of LLM work removes all race-related information from those narratives before presenting them to prosecutors for their initial consideration. This process takes one to two minutes.

Chohlas-Wood said the redaction goes well beyond explicit mentions of race and physical descriptions. It also takes out people's names and location information, including the names of neighborhoods or commercial establishment names, which he said are two of the biggest proxies for race.

"Knowing the location of an incident can be a very good hint as to the race of the people involved," he said. "So it's just a matter of trying to reduce the hints where they might pop up."

When the race-blind system was integrated into Yolo County's case management system, which processes 8,000 to 10,000 cases a year, Reisig said he didn't know what the data was going to show. A study of the data showed that there was no statistically significant variance in charging by race.

"Good, bad or ugly, we wanted to know," he said. "That was great news for us."

Regardless of what the data ultimately shows, Reisig said, the existence of a race-blind system in charging helps build the public's confidence.

"The public's trust in the criminal justice system has been greatly emboldened by this tool," he said.

From Pilot to Statewide Policy

Reisig, a career prosecutor who has served as the elected district attorney for 18 years, said concerns about racial bias in the criminal justice system have "dominated the conversation of my career."

In 2018, amid growing scrutiny around racial disparities in charging decisions, he began exploring a new approach: What if prosecutors could make charging decisions without ever knowing the race of the people involved?

The main question people sought answers about had to do with the way prosecutors handled cases involving non-white suspects, compared with their white counterparts.

"The most that a prosecutor could say was, 'Trust me,'" Reisig said. "And that's not good enough for a lot of people."

By the time Reisig integrated a race-blind charging process into the work of his office, the technology had already been studied in San Francisco.

In 2019, the city's then-district attorney, George Gascón, partnered with the Computational Policy Lab to experiment with an automated redaction tool that removed names, race and other identifiers from police reports for initial charging decisions.

The experiment, which was meant as a response to concerns of racial profiling, was short-lived. The San Francisco District Attorney's Office used a race-blind system for a few months, then stopped, never actually integrating it into its broader operations. The COVID-19 pandemic and changes in leadership hampered the development of a comprehensive system.

Gascón, a Democrat, resigned from the job in October 2019 and was elected the following year as Los Angeles County district attorney, only to face multiple recall campaigns targeting his progressive policies. Gascón's elected successor in San Francisco, Chesa Boudin, was recalled by voters in 2022.

Eager to try out the technology pioneered under Gascón, Reisig approached his connections at Stanford with the intent to build a redaction algorithm into the office's digital system. In 2021, the team created the program and offered it to the county prosecutor's office free of charge. In exchange, Computational Policy Lab researchers would have access to the data produced in the implementation. The data continues to be shared today.

"I told them I wanted to be the guinea pig," Reisig said.

Through the partnership with Stanford, Yolo County became the first district attorney's office to fully integrate the technology into its case management system.

At the time it began experimenting with race-blind charging, Yolo, a midsize county with a population of about 220,000 and about 40 prosecutors, was an ideal ground for a pilot project.

Its district attorney's office had already prioritized innovation, including the use of data via a public-facing transparency portal. It was among the first prosecutorial offices to lean on collaborative courts, for instance those addressing people with mental illnesses. It was also among the first DA's offices to ditch paper case files for digital ones.

"Initially, it was challenging, because it creates more work for people," Raven said. "But Jeff [Reisig] thought it was very important for us. And it wasn't done because he necessarily thought there was bias in charging cases. But he wanted to find out."

California legislators soon took notice, seeing Yolo County's pilot program as a potential blueprint for the state. The Yolo District Attorney's Office worked with California Attorney General Rob Bonta to draft legislation that would require all offices to start a race-blind charging program.

One year before race-blind charging rollout, the state attorney general's office issued a set of guidelines meant to guide district attorneys through implementation. These guidelines provide practical instructions for offices on how to redact reports — for instance what specific fields to black out — how to structure the two-step review, and how to document outcomes.

As the statewide mandate was going to take effect, some prosecutors were concerned that they would be in legal jeopardy for changing their charging decisions. But as the law went into effect, DAs found that they have leeway, in part because they can point to additional evidence they review at that second stage, Chohlas-Wood said.

In addition to providing the redaction software, the Computational Policy Lab is also involved in studying how the race-blind charging system impacts charging rates on a racial basis. A handful of district attorney's offices working with the lab have agreed to participate in the study by sharing its internal data on criminal charging. Findings are expected to be published, possibly next year.

The law requires prosecutors to share information on blind charging annually, which will enable larger-scale analysis. The California Department of Justice is expected to compile statewide data once the 2025 data can be harvested, probably well into 2026.

Meanwhile, jurisdictions outside California have also begun to experiment with race-blind charging.

Last year, Jackson County, Missouri, which includes Kansas City, began participating in a randomized controlled trial to evaluate race-blind charging. In partnership with researchers, Jackson County prosecutors are deploying an algorithm to redact race in case filings as part of a pilot program. Another pilot program is underway in King County, Washington, which includes Seattle, and it is also testing race-blind review procedures as part of a voluntary effort to reduce bias.

Funding and Technical Challenges

Chohlas-Wood said his lab's work has helped contain the costs associated with adopting a race-blind charging technology, which can vary widely, depending on which entity develops the software, as well as computing and staff costs.

"It's really important for us to make this as easy as possible, because we knew this could be otherwise very technically challenging or pretty expensive," he said. "The idea of actually raising money for something like this can be very hard for an office."

The Computational Policy Lab provides race-blind charging software to a majority of California counties — 34 out of 58 — including Los Angeles County, the most populous. Fourteen jurisdictions have hired private sector companies such as Sicuro Data Analytics and Meadowlark. District attorneys in Orange, Riverside, San Bernardino and Ventura counties have developed race-blind charging capabilities in-house.

The Computational Policy Lab develops and deploys its solution for free. District attorney's offices using Prosecutor by Karpel, a popular case management software, received the redaction upgrade at no cost. Another company, Journal Technologies, also offers a redaction application to clients using its platform, called eProsecutor, but charges fees for document processing and hosting costs, which amount to a few thousand dollars a year.

LeeAnn Karpel, the president of Karpel Solutions, said the company launched its Race Blind Charging Module in the fall of 2024 and offered training on how to use it before the California mandate went into effect.

"We developed this for our clients at no cost. Our California clients are currently using this module and processing their police report documents with no added fees," she told Law360 in an email.

The price tag involved in adopting a race-blind charging system is an obstacle for some jurisdictions in California, given that the state law mandating its implementation did not secure a funding stream.

Some prosecutors' offices have struggled to set aside IT staff for the integration process, in part because they have a backlog of technological issues they need to solve first. Another problem is time management, because the race-blind system essentially creates an additional step in the investigation process, it requires prosecutors to spend more time on their cases.

Anecdotally, district attorneys using the Computational Policy Lab solution have reported an increase of about 20% in the time spent on cases where blind charging was used. Some prosecutorial offices have said that they can't implement the new practice unless they can hire additional attorneys, but the state has not provided funding.

Reisig said that the concerns about costs voiced by some DAs are "very legitimate." When California legislators drafted the race-blind charging bill, the intention was for prosecutorial offices to be able to recover the costs from the state.

But as several district attorney's offices face financial constraints, the state has faced a budget shortfall of its own, leaving little sympathy for the counties seeking to adopt the new charging system.

"That's been a challenge," he said. "There has been a — I'll just say — a lack of understanding between the state and the counties on the implementation costs."

The district attorney's office in Riverside County, California, was another early adopter of a race-blind charging system. Unlike Yolo, Riverside developed it on its own by hiring app developers to build redaction capabilities from scratch, launching the system two years before the statewide mandate kicked in.

"We did it all in-house. And to me, it's been, it's been well worth it," Riverside County District Attorney Michael Hestrin, told Law360.

Building software internally, without having to rely on what he called a "cottage industry" fueled by the mandate, has saved the county money and allowed better customization, particularly in the context of data gathering on charging.

"Companies now are offering these off-the-shelf products, but they're very expensive," he said. "They're charging a lot to buy these things and then to maintain them."

There are also technical limitations to race-blind charging.

One of them is that there is a whole body of evidence — photos or videos, for instance — where racial indicators cannot be easily removed. Prosecutors have a choice between reviewing such evidence at the second stage of their investigation, following the initial race-blind charging decision. Or they can determine that the evidence is so crucial that a race-blind decision is just not possible, and therefore exclude the case from that process entirely.

Chohlas-Wood acknowledged that the inability to redact visual material could be a significant downside given the large and increasing availability of photo and video evidence produced by cellphones and surveillance cameras — but he also questioned the usefulness of such evidence in some cases.

"I think in certain case types that is going to be crucial, but in a lot of other case types, it doesn't seem like it's going to be that important," he said.

Because they involve more thorough and deliberate review of evidence, criminal investigations on serious crimes like murder or rape are less likely to be impacted by race-blind charging. On the other hand, Chohlas-Wood said, the blind approach is more useful in identifying unconscious bias in charging decisions that are made "casually and subconsciously."

Chohlas-Wood also cautioned that race-blind decisions made by prosecutors are not going to address larger disparity in arrests and other police actions, all of which happen upstream of charging decisions in criminal cases.

Pushback to Race-Blind Approach

In addition to concerns about funding, some prosecutors have shown reluctance to implement race-blind charging in part because they felt as if they were being led to admit they had an existing problem with bias. Still, a main reason A.B. 2778 passed unanimously in both the state's Senate and House of Representatives is that race blindness can have broad appeal across the political spectrum.

San Diego County District Attorney Summer Stephan, who serves as president of the National District Attorneys Association, said the association has not adopted a formal position on race-blind charging.

"The organization supports initiatives that aim to advance equal justice when they are thoughtfully designed, evidence-informed, and respect prosecutorial discretion," she said in an email. "In California, counties are in various stages of implementing the law, with some — like my office in San Diego County — already operational but others facing funding or technical hurdles."

Stephan added that it is too early to determine whether race-blind charging is helpful in detecting bias, or whether there is a widespread issue with prosecutorial bias at all, given that data collection is still underway.

"The preliminary outcomes show that prosecutors are making charging decisions based on the facts and evidence of the crime and not the race of the individual involved," she said. "However, it's too early to draw meaningful conclusions about outcomes."

Chohlas-Wood emphasized that while the Computational Policy Lab has devised its redaction system for prosecutors, it does not advocate for the practice on a policy basis.

"I want good outcomes," he said. "Next year, we might find that it doesn't have any impact on outcomes, and that it's just like there's no benefit to doing this. And that would be totally fine with me. Ultimately, what I'm after is understanding the impacts."

While justice reform advocates such as the American Civil Liberties Union have generally been supportive of the idea of taking race out of the equation, some public defenders argue that prosecutors should always factor in race when making charging decisions, but as a mitigating factor.

Brendon Woods, the Alameda County, California, chief public defender, told Law360 in an email that despite its stated goal to remove consideration of race from the charging process, race-blind charging does little to eradicate what he described as systemic racial disparities in the justice system.

"Race-blind charging sounds good on paper, but it's not going to remove racism that is woven into every layer off the criminal legal system," he said. "These cases that land on prosecutors' desks are still coming from overpoliced neighborhoods that are predominantly Black and brown. The reality is that prosecutors are eventually going to know the person's race, and it will lead to harsher sentences."

And Woods said California "is moving in the wrong direction."

In November, voters approved Proposition 36, which increased penalties for theft and repeat shoplifting.

As a result, Woods said prosecutors are seeking harsher punishments based on a person's past convictions. In Alameda County, which includes Oakland, 81% of the people represented by the public defender's office charged with offenses qualifying as felonies under Proposition 36 have been people of color.

"Let's not pretend that this is about resources. These same agencies that say they don't have the money to implement race-blind charging seem to have plenty of funding to prosecute people for so-called organized retail theft and drug offenses," Wood continued. "It is not about money, it is about priorities."

The San Francisco Public Defender's Office also expressed doubts on the impact of race-blind charging in combating bias in law enforcement.

"We have reasons to be skeptical," the office said in an email.

"Police are far more likely to stop and arrest Black and brown people than other races in our country and in San Francisco in particular," the office said. "That means that racial bias has already played a role in the arrests that land on a prosecutor's desk before any race-scrubbing software has kicked in."

Reisig acknowledged that race-blind charging doesn't solve all the problems that exist in the justice system.

"It doesn't satisfy everybody, but it's a good tool. It's a start," he said. "That's my view on it."

--Editing by Orlando Lorenzo.

Have a story idea for Access to Justice? Reach us at accesstojustice@law360.com. Access to Justice is taking a summer break. Our next issue will publish on Sept. 6.