Proportionality in AI ethics draws on a well-established legal and moral principle: that interventions affecting people's rights, freedoms, or opportunities must not go further than is necessary and justified. The proportionality principle presents a set of conditions - necessity, desirability, and suitability - that must be satisfied to justify the use of certain AI methods for a particular purpose.

Proportionality requires a rigorous assessment of three key factors, especially in high-impact settings like public services, security, employment, or emergency response:

• Suitability: Does the AI system help achieve the real, stated goal?

• Necessity: Is there a less intrusive way to achieve the same result without AI or with less invasive data?

• Balance: Do the benefits of the system outweigh the risks to privacy, autonomy, and civil liberties?

In AI, this often means avoiding “one-size-fits-all” automation and using stronger controls only where the risk is genuinely higher.

Why it matters

Proportionality matters because AI decisions can affect people’s rights, access to services, safety, and dignity. In a democratic society, it acts as a safeguard against technological overreach. 

Without it, there is a risk of "function or scope creep," where tools designed for extreme cases such as counter-terrorism are gradually redirected towards more mundane tasks like monitoring student attendance, and vice-versa, eroding the fundamental rights of citizens. 

Impacts & consequences

When proportionality breaks down, AI can become over-restrictive, unfairly punitive, or needlessly invasive. In automated decision-making, that can mean denying benefits, shutting down services, monitoring or flagging too many people, creating chilling effects, or using stronger interventions than the situation warrants.

In safety or security-sensitive uses, excessive reliance on automation can also transfer authority away from humans and create accountability gaps when things go wrong. Over time, this can reduce public trust and make legitimate AI use harder to defend.

Causes & contributing factors

Common sources and causes of proportionality failures in AI and automation include:

• Efficiency imperatives. Pressure to reduce costs and processing time leads organisations to automate decisions wholesale, without calibrating the scope of automated action to the severity of consequences.

• Lack of impact assessment. Deploying systems without prior human rights or proportionality impact assessments means potential harms are not weighed against intended benefits before rollout.

• Opacity and complexity: Algorithmic systems are often too opaque for decision-makers, courts, or citizens to evaluate whether their reach is justified.

• Data asymmetry. Systems trained on historical data that reflects prior over-enforcement or systemic bias inherit and amplify existing disproportionality against minority and low-income groups.

• Absence of proportionality review in procurement. Governments and corporations frequently purchase or deploy AI systems without requirements to demonstrate necessity or to audit scope against stated objectives.

• Regulatory gaps. Most jurisdictions lack binding proportionality requirements for AI, meaning systems can operate without legal accountability for overreach until harms are litigated.

• Automation bias. Decision-makers and oversight bodies defer to algorithmic outputs, reducing scrutiny of whether the system's reach is warranted in individual cases.