Artificial intelligence is undoubtedly one of the most remarkable achievements of human intelligence. Its contribution to medicine, education, scientific research, industry and everyday life offers possibilities that previous generations could scarcely have imagined. We should welcome these developments and recognise the efforts of those who have made them possible. Nevertheless, the greater the capabilities of technology, the greater our responsibility to ensure that it remains at the service of humanity.
During the recent 153rd Assembly of the Inter-Parliamentary Union in Arusha, Tanzania, I addressed the protection of children's rights in the age of artificial intelligence. I emphasised the responsibility of parliaments to ensure that appropriate legislation, impact assessments and accountability mechanisms are in place. Reflecting further upon this discussion, I believe our attention must now turn more decisively towards preventing harm rather than merely responding to its consequences.
A child cannot reasonably be expected to recognise every form of digital manipulation, nor should a parent be expected to understand every technological risk. The same applies to an elderly person exposed to deception, a patient relying upon technology in healthcare, or a worker whose livelihood may be affected by an automated decision. Even the engineers who develop these systems may one day find themselves exposed to their misuse. Protection must therefore be designed for everyone, with particular safeguards for those most vulnerable.
The question is not whether artificial intelligence can be made entirely free from risk. No responsible engineer could guarantee that. The question is whether we are taking every reasonable and technically achievable measure to prevent foreseeable harm, and whether those entrusted with public protection can verify that such measures remain effective. I believe the way forward rests upon six practical responsibilities.
1. Protection must begin within the system itself. Before carrying out sensitive operations, artificial intelligence should be designed to examine the action requested against established safety rules and the permissions granted. Where a prohibited operation can be reliably identified, it should be prevented rather than merely accompanied by a warning. This is particularly important where technology may be used to exploit children, obtain unauthorised personal information or interfere with systems affecting health, safety and financial security. The obligation should be proportionate to the risks involved and capable of being demonstrated through testing.
2. A system must not be left to judge its own safety. Independent technical controls should prevent it from acting outside its authorised functions or disregarding mandatory restrictions. These safeguards must be designed to resist attempts at circumvention, whether by an ordinary user or through the system's own behaviour. Developers should test them regularly, including after significant changes. Where serious weaknesses are identified, the affected functions should be restricted until appropriate corrective measures have been taken. An assurance that a system is safe should never replace evidence that its safeguards work.
3. When serious misuse is detected, prevention must be accompanied by an appropriate response. The system should be capable of refusing a prohibited operation and generating an internal safety alert. That alert should be assessed according to established procedures, with relevant evidence handled lawfully. Where legal reporting requirements are met, the competent authorities must be informed. European legislation already provides certain notification obligations, including for hosting providers becoming aware of specified suspected offences threatening life or safety and for serious incidents involving covered high-risk AI systems. Any wider reporting arrangement must be carefully defined by law. An automated suspicion cannot be treated as proof of criminal conduct, nor should protecting the public become a justification for indiscriminate surveillance.
4. Governments must ensure that these protections are not merely promises made by technology providers. The European Union has already established important obligations through its Artificial Intelligence Act, while Malta has designated the Malta Digital Innovation Authority and other competent bodies to undertake regulatory responsibilities. The next responsibility is effective implementation. Authorities must have the technical expertise, resources and lawful powers to examine compliance, require corrective action and impose appropriate measures where obligations are breached. For systems presenting serious risks, continued compliance should be demonstrable through documented testing, independent assessment where appropriate and verification that essential safeguards have not been improperly weakened or removed.
5. Responsibility must extend to those who develop, provide, operate and misuse artificial intelligence. Engineers and employees who identify serious dangers must have access to appropriate protected reporting channels. Providers and operators must remain answerable for their respective legal obligations, while individuals who deliberately use technology to commit unlawful acts must also be subject to the law. Responsibility should be established through proper investigation, not assumption. Where prevention fails, affected individuals must have access to complaints procedures and the remedies available under the applicable law.
6. International cooperation must make these safeguards more effective. Artificial intelligence crosses borders, while the responsibility for enforcement remains with competent authorities operating under national and regional laws. When a serious vulnerability is verified in one country, relevant information should be capable of reaching authorities elsewhere through lawful and secure arrangements. Common approaches to identifying serious incidents, communicating safety warnings and coordinating appropriate responses would reduce the risk of the same preventable harm occurring repeatedly. Existing institutions should be strengthened rather than unnecessarily duplicated.
These proposals do not suggest that existing legislation has failed or that governments have remained inactive. Important foundations are already in place. The challenge is to ensure that the obligations established by law translate into safeguards that operate effectively, can be independently examined and are strengthened as technology develops.
Nor should regulation discourage responsible innovation. Artificial intelligence offers considerable opportunities for researchers, smaller businesses, educational institutions and public services. Safeguards should reflect the risks presented by particular applications, protecting fundamental rights without placing unnecessary restrictions upon legitimate progress.
My experience in parliamentary life has taught me that the true value of legislation lies in its application. A right recognised in law must be capable of being exercised. A safeguard required by law must be capable of functioning. And an authority entrusted with public protection must possess the means to intervene when necessary.
The discussion in Arusha began, rightly, with children. Yet the responsibility extends to every generation. We cannot place the burden of protection solely upon those who may become victims, nor leave it entirely to the discretion of those developing the technology. Legislators, regulators, engineers and providers must each fulfil their responsibilities. Malta should seize this opportunity to lead by example, establishing a practical model of AI safety verification and enforcement that could serve as a reference for the wider international community.
Artificial intelligence should continue to advance. The question is whether our determination to protect humanity will advance alongside it.
Human intelligence created artificial intelligence. It must now devote the same ingenuity to ensuring that its safeguards are built into the technology, independently verified and upheld by the law. The real measure of progress will not be what artificial intelligence can accomplish alone, but what humanity has taken reasonable steps to prevent it from doing harm.
Dr Anglu Farrugia is Former Speaker of the House of Representatives of Malta