Recently, headlines have been calling for AI to be more regulated and to have more safeguards in place. Not only from people who are affected by it, but also from people helping to build it, who have publicly called for a slowdown, citing concerns about losing control of increasingly autonomous systems.
Irrespective of the real or supposed threat level, the technology has well and truly moved conversations on from science fiction to serious talks about regulation.
But what risks does it actually pose for our human rights?
To find out, I spoke to Elena Gaura, Professor of Pervasive Computing and Associate Pro-Vice-Chancellor for Research and Enterprise at Coventry University, about the rapid growth of AI and its effects on our human rights. Gaura has worked with AI, data, and intelligent systems for more than three decades, and her focus is on responsible and inclusive AI: bringing industry, researchers, educators, and government together so that the people affected by these technologies help to shape them.
Why is AI different from other technologies?
AI has scaled at an unprecedented rate. It no longer simply sends and retrieves data. It automates cognitive tasks. And, increasingly, it can now act on its own.
Gaura says AI adds three distinctive risk factors: speed, scale, and agency: “An AI system can produce or act on information at enormous scale, personalise its influence to an individual, and increasingly make or support decisions about employment, education, finance, healthcare, and public services … it interprets, predicts, recommends, and sometimes acts.”
It can also be hard to inspect: “If we do not know what data shaped a system, how it reached a conclusion, or who is responsible for using it, errors can spread before people even realise they have been affected.”
That doesn’t make AI inherently bad, she says, but it does mean that “the more powerful the technology becomes, the stronger our evidence, oversight, and accountability must become.”
AI didn’t create the risks. It amplified them
Generative AI has made some harms very visible, particularly deepfakes, impersonation, sexualised images, misinformation, and harmful interactions with vulnerable users. However, Gaura points out that the underlying risks aren’t new. Human-rights risks existed in algorithmic systems long before Generative AI arrived. What has changed is how fast and how far those risks can now travel.
According to Gaura, AI may affect:
- Privacy through surveillance and extensive data collection
- Equality through discriminatory recruitment, lending, or policing systems
- Freedom of expression through automated moderation
- Access to education, employment, healthcare, and public services
“More importantly, harm does not require malicious intent,” she says. “A system may discriminate because its data reflects historical inequality, because some communities were poorly represented during development, or because it is being used in a context for which it was never designed.”
That is why we should ask not only whether a system works on average, but “for whom does this work, who might it fail, and what happens to that person when it fails?”
Why are AI leaders calling for a slowdown?
Some AI company leaders have recently called for a slowdown, and Gaura says there is a distinction between stopping AI altogether, which is neither realistic nor necessarily desirable, and pausing a particular model, training run, or deployment when risks can’t be controlled. This, she says, is “simply responsible engineering.”
But why are they calling for a slowdown now? Gaura says it comes down to autonomy: “AI systems are becoming more autonomous and capable of longer, more consequential chains of action. Companies are encountering risks in practice, particularly around cybersecurity, control, and the security of their own development environments.”
There may also be commercial and strategic interests at play, and regulation can favour established companies. However, Gaura doesn’t speculate about individual motives. Her position is that “we should take the warnings seriously without allowing the companies making the technology to be the sole judges of the evidence.” The answer, she says, is independent evaluation, public transparency, and democratic oversight: “Safety cannot depend on whether a chief executive happens to exercise restraint.”
Who is accountable when it goes wrong?
“Responsibility may be shared, but it must never become so dispersed that nobody is accountable,” Gaura explains.
As Gaura sees it, developers are responsible for how systems are designed, tested, and documented. Organisations that deploy AI are responsible for deciding whether it is suitable and whether people remain meaningfully in control. Senior leaders are responsible for governance. Regulators and government must set enforceable boundaries.
However, Gaura says that for an ordinary person, the route to redress is fragmented. “Depending on the harm, they may need to approach the organisation concerned, or the Information Commissioner, an ombudsman, an equality body, a sector regulator, or ultimately a court. Many people will not know which route applies, and sometimes may not even know that AI influenced the decision.”
Gaura says a meaningful right to redress should include four things:
- Being told when AI materially affected a decision
- Receiving an understandable explanation
- Being able to obtain genuine human review
- Having access to correction, compensation, or another appropriate remedy
“A complaints button alone is not accountability,” she says.
Are certain people more exposed than others?
According to Gaura, certain demographics of people are more exposed than others, and that applies to both risks and benefits.
“Groups that were historically underrepresented in data may receive less accurate results. Groups that are already highly monitored may be overrepresented in surveillance or enforcement data. People with fewer resources, less digital confidence, or limited access to advice may also find it much harder to challenge an automated decision.”
Exclusion can work on the benefits side too. “If AI improves education, healthcare, and productivity, but only affluent organisations and communities can access it, it may widen existing inequalities.”
Gaura says that inclusion cannot mean asking people to comment after a system has been built. “Those who will live with the consequences must help define the problem, the acceptable trade-offs, and what success looks like. Civic trust is earned through participation, not persuasion.”
What should regulation look like?
Gaura says that regulation should be risk-based, enforceable, and adaptive, and focused on what a system can do, the context it’s used in, and the severity of possible harm, rather than treating “every spreadsheet or chatbot as though it were a frontier system.”
That means “there should be baseline expectations around transparency, data protection, security, and accountability” for everyone, with stronger duties for higher-risk applications. These include “impact assessments, testing for bias and foreseeable misuse, independent evaluation, incident reporting, traceability, human oversight, and a clear route to remedy. At the frontier, regulators must be able to delay or prevent deployment where developers cannot demonstrate adequate control.” Gaura also wants rules that don’t lock out universities and smaller companies, through sandboxes, shared testing facilities, and practical standards.
A single global legal regime is unlikely, she says. What is achievable is interoperability. “Shared definitions of serious risks, comparable testing, mutual recognition of credible assurance, cross-border incident reporting, and common minimum protections. International cooperation must include a broad range of countries, not merely those currently leading model development.”
It is Gaura’s hope that we can avoid a choice between innovation and rights altogether. “Trustworthy systems are more useful systems. They are more likely to be adopted, more resilient when circumstances change, and more capable of creating lasting public value.”
Yet, this optimism comes with a condition. “People must remain authors of the future, not merely consumers of technologies designed elsewhere.”
Whether or not people’s worst fears about AI ever come true, don’t wait for science fiction to become fact before acting. AI harms are already here. Opaque decisions. Unequal outcomes. No clear route to redress. But as Gaura points out, there are simple steps to follow: knowing when AI affected a decision, getting an explanation, and being able to ask a human to review it. The tools to tackle these barriers are becoming within reach. So the real question isn’t whether AI will turn on us. It is whether we will stay in charge of it.
