Artificial intelligence is already shaping real decisions that affect people. This impacts what content is removed online, how harmful behavior is reported, and how public authorities manage risk.
In many cases, AI is no longer something that is quietly tested in the background. It is already part of the way platforms and institutions operate.
For many years, responsible AI has been discussed primarily as an ethical issue. We talked about fairness, bias, transparency, and values. These conversations mattered, and they still matter.
But many of the AI failures we see today are not just due to ethical inconsistencies or technical flaws. This happens because responsibilities are unclear, oversight is weak, and decision-making power is spread among too many stakeholders.
In other words, AI has become a governance issue.
When AI systems fail, governance is usually the first to fail.
Many countries are now using AI to manage scale. Social media platforms rely on automated systems that process millions of posts every day. Public authorities use AI tools to prioritize cases, monitor online victimization, and support enforcement operations.
When something goes wrong, the first question often asked is whether the model is accurate enough. That question misses the deeper issue. In many cases, the technology could have worked better, but the governance around it failed.
Common governance gaps include:
• There is no clear owner of the AI system.
• Limited pre-deployment monitoring.
• Weak escalation when harm begins to appear.
• Responsibility is divided between those who build the system, those who deploy the system, and those who are expected to regulate the system.
These gaps are well recognized in international policy discussions on AI accountability, including efforts by the OECD and the WEF AI Governance Alliance.
Lessons learned from online harm and content management
Many of these challenges were discussed in recent papers. Future conversations In our podcast on hate speech, deepfakes, and online safety, researchers and regulators speak candidly about the limits of AI and regulation in practice.
One message was clear. AI moderation tools already exist and are widely used. Machine learning is essential as a first filter for harmful content. A more difficult issue lies in how these tools are managed.
Content moderation typically works in layers.
• Automated system warns of potential harm
• Human moderators review complex or pending cases
• Regulators intervene if the platform does not function.
When accountability is lacking in one or more of these layers, dysfunction occurs. Platforms may lack investment in local language and cultural backgrounds. Surveillance may rely on complaints rather than prevention. Responsibility may shift between the companies building the systems, the platforms deploying them, and the authorities expected to oversee them.
These weaknesses are even more pronounced in multilingual and culturally diverse societies. Language mix, slang, and context change quickly. Without strong governance, even capable AI systems will struggle to keep up.
Where responsibility breaks down
This network helps explain why AI damage is rarely caused by a single failure. AI systems are developed by one group, deployed by another, monitored at a distance, and most directly experienced by the public.
When ownership, oversight, and escalation are not clearly linked, the interstices between organizations become harmful.
Child safety shows why governance matters most
The risks are especially clear when children are involved. Deepfakes and synthetic images generated by AI have made online exploits easier to create and harder to detect. UNICEF has warned that AI poses new risks to children that cannot be addressed by technology alone.
A recent example clearly illustrates this. In January 2026, Grok, a chatbot associated with X, came under intense global scrutiny after it was reported that it could be exploited to create non-consensual sexual images, including sexual images involving minors. Malaysia reported a detailed description of the incident and wider risks. bernama.
This is important because it shows how quickly damage can move from niche tools to mainstream platforms. Features that should be blocked by design can spread far and wide before safety measures can catch up.
It’s not just a failure of detection, it’s a failure of governance.
In many countries, such content is already prohibited by law. Australia’s online safety framework is set out in the Australian Government’s Legislative Summary and is enforced through the powers set out by the eSafety Commissioner.
In Malaysia, the Online Safety Act 2025 came into effect on January 1, 2026, and supporting legislation and FAQs have been released to coincide with its implementation. But enforcement remains difficult when platforms operate across borders and spread harmful substances faster than regulators can respond.
These examples show that child safety is not just a technology issue. It’s a governance issue.
AI in the public sector comes with hidden risks
AI is also being deployed across public services, from education and welfare to digital enforcement and online safety. These systems affect real outcomes for real people.
When public sector AI fails, the impact extends beyond technical performance. It affects trust in the organization.
However, governance often lags behind implementation. AI systems may be introduced without independent review, clear accountability for outcomes, and transparent ways for the public to question decisions. When something goes wrong, a simple question arises:
Who is responsible?
If institutions cannot answer these questions clearly, public trust will rapidly erode.
What does responsible AI actually look like?
Responsible AI does not mean avoiding AI. It means governing it properly.
In practice, this includes:
• Clear ownership of each AI system.
• Defined roles for monitoring and review.
• Documented decision making and risk assessment
• Continuously monitor real-world impact
• Ability to suspend or withdraw the system in case of harm.
It also means recognizing that not all risks can be solved by a better model. Decisions regarding permissible use, escalation, and enforcement are made using human judgment. Leadership is required at senior and board level.
Across many jurisdictions, regulatory expectations are already changing. Online safety laws, platform mandates, and public sector guidelines demonstrate that responsible AI is moving from voluntary principles to enforced governance.
From discussion to decision making
Responsible AI has moved from discussion to decision-making.
Important questions are no longer abstract.
• Who owns the system?
• Who will oversee it?
• Who will act if harm begins to appear?
Institutions that cannot clearly answer these questions will face regulatory, reputational, and trust risks, no matter how much technology advances.
As AI becomes more integrated into public life, responsible AI must be treated as a core responsibility of governance. In this way, trust is built, harm is mitigated, and innovation can continue in a way that society accepts.
This article first appeared on Monash Lens. Read original article
