Building safer AI – A responsibility-driven approach

Article: June 2025

Building safer AI – a responsibility-driven approach

By Huw Bristow, Chief Technology Officer, Altia

In an age where artificial intelligence (AI) is transforming industries and reshaping the way we live and work, ensuring its safe and ethical use has never been more pressing. For those working in the police and law enforcement sector, the stakes are particularly high.

AI has vast potential to revolutionise investigations, making them more streamlined to alleviate workloads, deliver time efficiencies, and identify avenues of interest which would otherwise go unnoticed by the naked eye. But there are risks if it is not implemented responsibly.

The relationship between police forces and the public can be inherently fragile and the introduction of AI technologies, if mishandled or perceived as intrusive, risks exacerbating this. Therefore, AI must be implemented thoughtfully and responsibly.

Identifying the risks

One of the most pressing challenges lies in identifying and mitigating ethical risks. AI systems are only as unbiased as the data they are trained on, and skewed or incomplete datasets can lead to discriminatory outcomes.

In investigative contexts, where the consequences of bias can profoundly impact individuals and communities, transparency and accountability has never been more in important in AI deployment. Decisions driven by algorithms need to be explained and justified, otherwise the trust placed in these systems and the institutions that use them can quickly erode.

Additionally, as AI often relies on vast amounts of data, there is an inherent tension between leveraging data to enhance effectiveness and protecting an individuals’ privacy. Striking the right balance is essential to avoid the unintended consequence of data misuse or breaches. Making a decision about whether to allow your data to be included in AI training is a difficult choice, and in most cases would be inappropriate for law enforcement, but this doesn’t reduce the opportunity to get the benefits of AI more widely.

Meanwhile, operational risks pose a challenge of its own. While AI has the power to enhance efficiency and accuracy, over-reliance on it can lead to the deskilling of human operators and devaluing the role of human judgement and intuition in investigations. Trusting AI outputs without critical evaluation can derail justice, leading to wrongful accusations or missed opportunities to hold perpetrators accountable.

And then, legal and regulatory landscapes add another layer of complexity. Compliance with frameworks such as GDPR, the EU AI Act, and local government policies is non-negotiable, yet navigating these regulations can be a labyrinthine process.

Mitigating the risks

While public perception can quickly turn negative, and any high-profile missteps in the use of AI can attract intense media scrutiny, the rewards outweigh the risks if government agencies and their partners utilise AI responsibly.

It begins with a multi-faceted approach. Ethical risks can be addressed by implementing thorough auditing processes to identify and rectify biases in datasets and algorithms. Ensuring diversity in the team designing AI systems can help mitigate the risk of unconscious bias and, by embedding transparency mechanisms, such as explainable AI, on premise AI developers can enhance accountability by making the decision-making process of AI systems more understandable for end-users.

Training programs for operators can ensure that users of AI maintain critical thinking skills and remain vigilant against over-reliance on automated outputs with a view that AI is a tool to aid the human mind, not replace it. Establishing fail-safes and maintaining high standards of data quality can enhance the accuracy and reliability of AI outputs. By emphasising the complementary role of humans and AI, organisations can avoid the pitfalls of deskilling while leveraging the strengths of both.

Addressing legal and regulatory risks involves staying ahead of evolving compliance requirements and fostering collaboration with regulators. Regular audits, clear documentation of AI processes, and robust legal counsel can help organisations navigate legal frameworks. Meanwhile, building partnerships with industry bodies can contribute to the development of best practices and standards.

Seizing the opportunity

AI is revolutionising policing by streamlining operations and enhancing efficiency. Tools such as generative AI and deep learning software interpret vast datasets, transforming tasks including automating case management and disclosure tasks, identifying and alerting investigations to anomalies in large datasets, interview transcription, and entity matching. Real-time transcription extracts critical insights quickly, while summarisation tools condense large textual data for rapid decision-making in live crime situations.

The technology enhances transparency and reduces administrative burdens by automating audit trails and tracking decision processes too. In short, AI maximises value, speed, and usability in policing.

As police forces look to utilise AI, they must engage with the public, media, and other stakeholders to affirm its responsible and ethical use, encouraging a culture of openness and accountability. This is crucial to tackling public scrutiny.

By prioritising transparency, fostering robust data practices, and integrating human oversight into AI systems, developers can mitigate risks and ensure the technology serves as a force for good. In a rapidly evolving field, this approach will see us harness the transformative power of AI while safeguarding the principles that underpin justice and trust in our society.

Huw Bristow is Chief Technology Officer at Altia, a Nottingham-based software business which creates intelligence and investigation software for law enforcement agencies, government departments and private sector businesses. Altia has been a specialist in this area for almost three decades, with offices also based in Glasgow.