From laboratory to legislation: Why AI equity is a safety requirement

We must stop treating AI equity as a social preference. It is a fundamental safety requirement and it is time we legislated accordingly.  

The intersection of science and social justice  

My perspective is shaped by several overlapping lenses: the rigour of biomedical science, the delivery principles of agile technology, the long view of education, and the systemic reach of policy. When these lenses converge, they reveal the same truth: technological bias is a social justice issue disguised as a technical oversight.  

In the laboratory, we use controls to ensure accuracy. We build in safeguards, run parallel checks, and validate every step before a result reaches a clinician. In society, our controls are the diverse voices at the leadership table. When those voices are absent, we lose our most important quality assurance mechanism. The resulting technology does not just underperform, it actively excludes.  

This is why, in orchestrating the proposal for the APPG's 2026-27 project, my goal was to move the conversation beyond abstract ethics. Ethics without accountability is just aspiration. We need technical standards, legislative levers, and institutional change.  

The agile approach to equity  

In the technology sector, we talk constantly about ‘agile delivery’, moving fast, iterating, shipping. But speed without safety is a hazard, not a virtue. I have seen this in the lab, and I see it now in how AI is being rolled out across public and private institutions.  

As a Scrum Master - think of this like a coach for an agile team in tech - I think about equity in terms of a ‘definition of done’: a product is not finished if it contains systemic blind spots that harm protected groups. It does not matter how elegant the code is, or how impressive the performance metrics look on paper. If it fails the people it is supposed to serve, it is not done.  

We need to shift from a reactive culture, discovering bias after harm has occurred, then scrambling to patch it, to a proactive culture of equity by design. This is not a values statement. It is an engineering discipline. It requires building representative datasets, testing across demographic groups as a standard protocol, and ensuring that the teams writing code have the lived experience to spot the gaps before they become headlines.  

The gap between where we are and where we need to be is not a mystery. It is a choice.  

Gendered harm is not a glitch  

Let me be direct about something that is too often softened in policy discussions: the harms caused by biased AI are not accidents. They are the predictable consequence of building systems without the people those systems affect.  

When an algorithm fails to identify a female cardiovascular event because it was trained predominantly on male physiology, that is a failure of science.  

When a recruitment tool filters women out of senior STEM roles because it learned from decades of historically male hiring patterns, that is a failure of justice.  

When a risk assessment tool in the criminal justice system performs differently across ethnic groups because the training data encoded existing inequalities, that is a failure of governance.  

None of these are rare or extreme cases, they are documented, published, and in some instances already subject to legal challenge. What we lack is a systematic legislative framework that treats these failures with the seriousness they deserve. Gendered harm in AI is not a glitch. It is a failure of leadership and a clinical oversight, and we need to start calling it both. 

My hope with the new APPG project is that it will not simply catalogue failures, but will identify the mechanisms of change - specific, actionable interventions that can move us from the status quo to a system that is safe for everyone.  

A call to action for the STEM community  

The transition from the laboratory bench to the halls of Westminster has reinforced a belief I have held throughout my career: representation is our most effective form of risk mitigation. Diverse teams do not just produce more equitable technology. They produce safer technology. They catch what homogeneous teams miss. They ask the questions that never get asked when everyone in the room shares the same background.  

But I also know, as an educator and mentor, that we cannot change the system without changing who is in the room and that takes deliberate, sustained effort at every level, from school curricula to boardroom appointments.  

So, I am calling on colleagues across the STEM spectrum, including researchers, data scientists, clinicians, engineers, and policymakers to engage with the APPG’s project. We need your evidence. We need your expertise. We need your lived experiences, because they are not a soft addition to this process. They are the most rigorous data we have.  

Inclusion is not about filling a quota. It is about ensuring that the technology we build tomorrow is safe enough and fair enough for everyone to use today.  

Let us move beyond the shadows of bias and build a future where equity is not an aspiration. It is the baseline.

Image: Female software engineer codes on a laptop © This is Engineering 

Guest blog authors are invited to write for the British Science Association about subjects which align with our vision and mission. The views expressed in guest blog posts are those of the author and do not necessarily reflect the official stance of the BSA.