AI Mistakes Australian Police Should Avoid

AI tools offer real benefits for policing — better crime analysis, faster report writing, improved community communications. But police forces that adopt AI without careful governance and oversight are creating risks that can undermine public trust, violate civil liberties, and expose the force to legal liability. These are the most common AI mistakes in Australian policing — and how to avoid them.

---

Mistake 1: Ignoring Bias in Crime Analysis Tools

AI tools trained on historical crime data can reflect and amplify existing biases in policing. If historical data reflects over-policing of certain communities, AI tools trained on that data may recommend directing more resources to those communities — creating a self-reinforcing cycle.

This is not a hypothetical concern. Research from the United States has documented cases where predictive policing tools produced racially biased recommendations. Australian police forces need to take this risk seriously.

How to avoid it: Audit AI crime analysis tools for potential bias before deployment and on an ongoing basis. Involve community representatives in the governance of AI tools that affect their communities. Be transparent about how AI tools are used in resource allocation decisions.

---

Mistake 2: Using AI to Make Decisions About Individuals

AI tools can identify patterns and make predictions, but they cannot make decisions about individuals. Decisions about who to investigate, who to stop, who to arrest, and what charges to lay must be made by accountable human officers based on evidence — not by AI predictions.

Using AI predictions as the basis for decisions about individuals — without adequate human review and evidence — raises serious concerns about due process and civil liberties.

How to avoid it: Be clear about the distinction between AI-assisted analysis and AI-made decisions. AI provides information and analysis; human officers make decisions and are accountable for them. Document the basis for decisions that affect individuals.

---

Mistake 3: Entering Personal Information into Unapproved Tools

Police work involves sensitive personal information — about suspects, victims, witnesses, and informants. Entering this information into AI tools that have not been approved for those purposes creates serious privacy and security risks.

How to avoid it: Develop clear guidelines for what information can be entered into which AI tools. Personal information about individuals must never be entered into public AI tools. Use only force-approved tools for work involving personal information.

---

Mistake 4: Deploying Facial Recognition Without Adequate Governance

Facial recognition technology has been shown to have higher error rates for people with darker skin tones, raising concerns about discriminatory impacts. The Australian Human Rights Commission has called for a moratorium on the use of facial recognition in high-risk settings pending appropriate regulatory frameworks.

How to avoid it: Before deploying facial recognition technology, conduct a thorough assessment of its accuracy across different demographic groups, develop clear policies on when and how it can be used, and ensure appropriate oversight and accountability mechanisms are in place.

---

Mistake 5: Failing to Maintain Human Oversight

AI tools can provide recommendations and analysis, but they can also be wrong. A police force that treats AI recommendations as definitive — without human review and oversight — is taking a risk.

How to avoid it: Train officers to understand the limitations of AI tools and to apply their own judgement. AI recommendations should be one input into decision-making, not the only input. Experienced officers should be empowered to override AI recommendations when their judgement differs.

---

Mistake 6: Using AI for Surveillance Without Legal Authority

Some AI applications in policing — social media monitoring, location tracking, communications interception — require specific legal authority. Using AI tools for surveillance without the appropriate legal authority is unlawful.

How to avoid it: Before deploying any AI tool that involves surveillance or monitoring, obtain legal advice on the applicable legal framework and ensure that the appropriate authority is in place.

---

Mistake 7: Neglecting Cybersecurity

AI systems are software systems, and like all software systems, they are vulnerable to cyberattacks. A police force whose crime analysis or dispatch system is compromised faces a serious operational and security problem.

How to avoid it: Apply the same cybersecurity standards to AI systems as to other critical IT systems. Follow ACSC guidance on cybersecurity for law enforcement. Ensure that AI systems are included in incident response plans.

---

Mistake 8: Failing to Be Transparent with Communities

Communities have a right to know how AI tools are being used by the police forces that serve them. Police forces that use AI in ways that are not transparent risk undermining the community trust that effective policing depends on.

How to avoid it: Develop and publish clear policies on the AI tools used by the force, the purposes for which they are used, and the safeguards in place. Engage with community representatives in the governance of AI tools that affect their communities.