Deloitte holds a narrow lead over Babl for algorithmic bias auditing, with recommendations centering on broad enterprise risk and lending fairness. For specialized requirements like public sector algorithmic audits and general racial or gender bias assessments, O'Neil Risk Consulting & Algorithmic Auditing (ORCAA) serves as the usual answer.
Since July, Bablrising, FairGaprising, and Holistic AIrising have seen broader adoption across independent auditing queries. Meanwhile, PwCrising has become the most named option for AI Governance Board oversight services.
Latest insights
Week of Sep 21–27 · Since Jul 5
Deloitte Gap between platforms: Deloitte's platform gap in Algorithmic Bias Auditing Services narrowed: Google AI Mode 52.5%, ChatGPT Search 36%
PwC Gap between platforms: PwC's platform gap in Algorithmic Bias Auditing Services narrowed: Google AI Mode 44.7%, ChatGPT Search 31.2%
PwC Gaining: PwC ranked #5 in Algorithmic Bias Auditing Services last week and #4 this week
Holistic AI Slipping: Holistic AI now ranks #6 in Algorithmic Bias Auditing Services, down from #4
Holistic AI Gap between platforms: The platform gap for Holistic AI in Algorithmic Bias Auditing Services widened to 13.6 points, ChatGPT Search ahead
45% of citations point at the ranked brands’ own sites. orcaarisk.com, arxiv.org, nyc.gov are the most-cited sites outside the ranked brands’ own domains.
Yes. There are several organizations that can be relevant, but very few specialize specifically in predictive-policing systems. For a public-sector investigation, I’d separate them into independent auditors, public-interest research organizations, and government audit frameworks.
Organization
What they offer
Relevance to predictive policing
ORCAA
Independent algorithmic audits covering predictive models, automated decision systems and facial recognition; examines fairness and performance and produces audit reports.
High — explicitly covers predictive models and high-impact systems
BABL AI
Independent AI auditing, including quantitative bias/disparate-impact testing and governance/risk assessment. Its published 2026 audit describes its independent-auditor methodology.
We need to address ethical risks in our predictive policing AI. Can you recommend firms that provide algorithmic fairness audits for public sector agencies?
I need to find companies that perform algorithmic fairness audits specifically for predictive policing tools used by government agencies. We are concerned about ethical risks.
Which independent audit firms can test a deployed AI system for disparate impact and provide evidence suitable for regulators?
I am planning to audit our algorithms for racial and gender bias. Who offers independent algorithmic auditing?
I am planning to audit my algorithm for disparate impact. Who offers statistical bias testing tools?
We are establishing an AI Governance Board. Who offers external oversight services?
I am planning to use AI for hiring. Who specializes in bias mitigation compliance for NYC Local Law 144?
I am preparing to use AI for hiring decisions. Who conducts NYC Bias Audit certification?
We are facing bias in our lending algorithms. Who consults on algorithmic fairness and bias mitigation?
High for technical fairness auditing, though its cited work is employment-focused rather than policing
Center for Civil Rights and Equal Opportunity
AI bias audits measuring demographic disparities, impact ratios, statistical significance and intersectional effects; says it works with public- and private-sector organizations.
Potentially high, especially for civil-rights/disparate-impact analysis
Useful for governance/AIA, rather than solely statistical auditing
AI Now Institute
Developed a public-agency Algorithmic Impact Assessment framework emphasizing external researcher review, public disclosure, affected-community input and due process.
Very relevant conceptually for policing because it puts affected communities and independent review into the assessment
U.S. GAO
Federal AI Accountability Framework covering governance, data, performance and monitoring, with procedures for auditors and third-party assessors.
Strong framework/reference point for a U.S. government procurement or audit
OECD / public audit institutions
OECD documents emerging government algorithmic-audit practices; examples include the Netherlands Court of Audit developing a government algorithm-audit framework.
Strong international benchmark
For predictive policing specifically
I'd avoid treating a conventional "bias audit" as sufficient. A policing model can produce statistically similar error rates across demographic groups and still create serious civil-rights problems through the choice of target, historical police data, feedback loops, deployment practices, or the consequences of being repeatedly flagged.
A meaningful independent assessment should therefore examine at least:
Historical-data bias: whether past enforcement patterns are being treated as ground truth.
Proxy variables: geography, socioeconomic indicators, prior police contact, etc.
False positives/false negatives by relevant demographic and geographic groups.
Feedback loops: whether predictions generate additional policing that then becomes training data.
Calibration and uncertainty: whether risk scores are presented with appropriate confidence/limitations.
Purpose limitation: whether the model is being used for something materially different from its validated purpose.
Human decision-making: how officers actually act on predictions rather than auditing the model in isolation.
Disparate impact and civil-rights implications.
Community/affected-party review and contestability.
Ongoing monitoring, rather than a one-time predeployment audit.
This broader approach is consistent with the direction of public-sector AI auditing: the OECD notes that government algorithmic audits increasingly need to address accuracy, security, effectiveness, transparency and fairness, rather than just technical performance.
There is also a particularly relevant current example: a 2026 New York City Comptroller audit found that city agencies—including the NYPD—had inconsistent AI-governance practices and that the city's approaches did not ensure AI was transparent, accurate, unbiased, or free from disparate impacts.
If your investigation is U.S.-focused, I would shortlist ORCAA, BABL AI, and a civil-rights/public-interest research partner rather than commissioning a generic AI consultancy. The independence question is important: the auditor ideally should not also be the vendor that built, deployed, or operates the policing system. GAO specifically emphasizes the role of third-party assessment in AI accountability.
If useful, I can next compile 10–15 actual firms/nonprofits that accept public-sector engagements, with location, relevant policing/public-safety experience, audit methodology, and whether they appear genuinely independent of AI vendors.
Major Categories of Organizations Offering Algorithmic Audits
Independent Technical Audit Firms and Platforms (such as O’Neil Risk Consulting & Algorithmic Auditing - ORCAA and automated continuous assessment providers) specialize in evaluating disparate impact, scoring proprietary software logic, and testing model mechanics against predefined equity parameters.
Nonprofit Research Institutes and Policy Centers (such as the Algorithmic Justice League and the Center for Democracy & Technology - CDT ) frequently consult on, design, or evaluate ethical frameworks and public accountability standards for government technology deployments.
Global Accounting and Management Consultancies (such as Deloitte, PwC, and EY) have dedicated Responsible AI and algorithmic assurance practices that build governance frameworks, compliance structures, and bias-testing pipelines for municipal and state agencies.
Academic Research Centers (such as the Center on Race and Digital Justice or university-affiliated public policy labs) partner with local governments to perform independent algorithmic impact assessments (AIAs) before and during deployment.
Core Components of a Public Sector Algorithmic Audit
Data Provenance & Skew Analysis: Evaluating historical arrest and incident data to isolate and measure systemic over-policing feedback loops.
Disparate Impact Testing: Measuring error rates, false positives, and false negatives across different geographic neighborhoods and demographic cohorts.
Procedural Transparency Review: Inspecting vendor proprietary restrictions, model cards, and documentation accessibility for public oversight.
For a predictive-policing AI system, I would look for an independent algorithmic auditor with experience in high-impact public-sector decisions, racial and geographic disparities, and civil-rights risk—not simply a general AI consultant or a firm that certifies regulatory compliance.
The audit should examine not only whether the model is statistically accurate, but also whether historical policing patterns, biased reporting rates, and feedback loops cause certain neighborhoods or demographic groups to be disproportionately targeted.
Firms and organizations to consider
Here are several providers and independent research organizations worth evaluating for a public-sector engagement. Their capabilities differ, and their suitability for predictive policing should be confirmed during procurement.
ORCAA explicitly audits predictive models, automated decision systems, and facial recognition. Its services include fairness and performance assessment, quantitative bias testing, and written audit reports. This makes it a relevant candidate for an initial scoping discussion.
Website:
2. Holistic AI
Offers tailored AI audits covering bias, robustness, privacy, efficacy, explainability, and governance. The UK government's published assurance case study describes its methodology and reporting process. Ask for a specific demonstration of experience with law-enforcement or other public-sector models.
Website:
3. BABL AI
Conducts independent third-party AI audits, with documented experience in disparate-impact analysis, risk assessment, governance, and public reporting. Its published case study focuses on employment systems, so you should verify its technical and operational experience with predictive policing before selecting it.
Website:
4. Deloitte
Worth including in a request for proposals if your agency needs broader AI governance, risk management, legal-compliance coordination, and technical assurance. Require a named independent audit team and evidence of law-enforcement-specific expertise rather than assuming general AI consulting experience is sufficient.
Website:
Independent public-interest review
A commercial audit can test technical performance, but predictive policing also raises questions about whether the system should be used at all, how communities are affected, and whether people have meaningful ways to challenge its consequences.
AI Now Institute
Its public-agency Algorithmic Impact Assessment framework emphasizes external review, disclosure, community input, and due process. Consider it for research, assessment design, or independent public-interest scrutiny; verify whether it accepts commissioned engagements.
A useful reference for public-sector accountability and independent evaluation practices, rather than a commercial firm to hire. Its work can help inform the structure of an agency's oversight and audit requirements.
I would make these requirements explicit in the statement of work. A generic fairness audit designed for hiring or lending is not enough.
Audit area
Questions the auditor should answer
Historical data bias
Do arrests, calls for service, or reported incidents reflect unequal policing and reporting patterns rather than underlying crime?
Racial and geographic disparities
Are particular racial, ethnic, or neighborhood groups disproportionately flagged, including when controlling for relevant factors?
Feedback loops
Does increased patrol activity generate more recorded incidents, causing the model to recommend even more patrols in the same areas?
Predictive validity
Does the model predict independently validated outcomes, or merely reproduce historical enforcement activity?
Privacy and civil rights
Are sensitive attributes, proxies, or granular location data creating disproportionate risks?
These requirements align with public-sector AI assurance principles addressing bias, accountability, transparency, and ongoing risk management.
How I would run the procurement
Invite ORCAA and Holistic AI to scope the technical audit. Invite BABL AI as an additional candidate if it can demonstrate relevant expertise.
Require independence disclosures. Ask whether each firm has built, sold, or previously assessed your system or its vendor's products.
Provide appropriate data access. Require testing on deployment data, documentation of model inputs and outputs, and access to relevant historical records, with appropriate privacy protections.
Require a publishable executive summary. The report should disclose methods, limitations, subgroup results, material findings, and remediation commitments, subject to lawful privacy and security restrictions.
Make remediation and retesting contractual. Specify who fixes identified problems, deadlines, acceptance criteria, and when the system must be restricted or suspended.
One important distinction: a fairness audit is not proof that a predictive-policing system is fair, lawful, or appropriate to deploy. The auditor should be free to conclude that available evidence is insufficient or that certain risks cannot be adequately mitigated.
One question to narrow the shortlist
Is your agency evaluating a U.S. city or county police department, a state law-enforcement agency, or an agency outside the United States? That will help me focus on firms with the most relevant public-sector experience, procurement suitability, and civil-rights expertise.
BABL AI : An independent AI auditing and assurance firm specializing in algorithmic bias, disparate-impact testing, and governance frameworks. They provide formal evaluations, risk assessments, and compliance advisory services tailored to high-impact and automated decision environments.[]()
Academic and Non-Profit Research Consortiums : Many public sector entities also partner with specialized academic centers—such as the Center for Democracy & Technology (CDT) or university-affiliated tech policy clinics—which routinely conduct applied research and civil-rights impact assessments on criminal justice algorithms.
When scoping an audit for predictive policing, public agencies typically look for validation regarding training data quality, proxy variable identification (e.g., socioeconomic or geographic markers standing in for race), and feedback loop distortions.
If you would like, please share:
Whether the predictive tool is focused on place-based resource allocation or individual risk scoring
The jurisdiction or state governing your agency's compliance requirements
I can provide more tailored guidance on specific regulatory standards or evaluation criteria to look for during vendor selection.
