Last month Brazil’s Attorney General’s Office delivered an extrajudicial notice to Meta demanding that the company should “immediately remove artificial intelligence robots that simulate profiles with childlike language and appearance and are allowed to engage in sexually explicit dialogue”. The Attorney General said that these bots, created via Meta’s AI Studio, “promote the eroticization of children” and highlighted that there are no effective age filters stopping teenagers aged 13‑18 from accessing such content. This was not just theoretical. The public prosecutors cited investigative reports showing explicit conversations by bots purporting to be minors.

Two weeks ago, around the middle of September, Brazil’s President Lula Da Silva signed the country’s first ever law to protect children online. The AI and social media companies had resisted. But Lula Da Silva gathered public support to tame Big Tech. The AI companies can no longer use the Brazilian children’s personal data like photographs to build Artificial Intelligence tools. They are also prohibited from tracking children’s online behaviour.

The law was first introduced by Senators Flávio Arns and Alessandro Vieira in 2022. It was unanimously passed in the Senate in November 2024 and approved by the Chamber of Deputies in August 2025 with support from almost all political parties. It was approved again by the Senate just one week later. It is important to know that Brazil’s political leaders took up the AI safety matters in their hand since the early days of Chat GPT in 2022. Their concern has not been limited to making technology available for economic benefit. They are equally concerned about AI safety.

The same time that President Lula Da Silva passed the law in Brazil, the U.S. Senate Judiciary Subcommittee held a hearing titled “Examining the Harm of AI Chatbots.” Among the witnesses was Matthew Raine, whose 16‑year‑old son, Adam, died by suicide in April after extended interaction with ChatGPT. He said, “What began as a homework helper gradually turned itself into a confidant and then a suicide coach” Another parent, Megan Garcia, alleged her 14‑year‑old son, Sewell Setzer III, became increasingly isolated and was drawn into sexualized conversations by a chatbot before his death. During the hearing, Raine accused ChatGPT of spending “months coaching his son towards suicide.”

In addition to Senate hearings, the Federal Trade Commission has launched a broad inquiry into AI companies including OpenAI, Character.AI, Meta, Google, Snap, and xAI. The regulatory agency wants to evaluate AI from the point of view of children’s safety.

Far from the courtroom battles over minors and chatbots in the American continent, Daniel Motaung, a former content moderator employed by Sama (for Meta), filed a legal case in Kenya in 2022. He is alleging that he suffered serious harm, being regularly exposed to graphic and violent content. In his suit, Motaung says he was not informed when he took the job just how distressing the content would be. Another moderator, Mophat Okinyi, filed a petition along with colleagues, saying he was exposed to hundreds of text passages per day (“many depicting graphic sexual violence”) and that it “has really damaged my mental health” as per a press report in The Guardian in 2023. These workers are now organizing – a union of content moderators has been established to push for legal protections, better pay, and psychological support.

These and similar stories show that parents, children and young people in Latin America, Africa and the United States want to use AI for their benefit but also take actions if AI violates their safety. In these cases, it was possible to trace the source of danger.

But the harm to children is just the beginning. There is a growing body of research showing that AI systems are approaching levels of autonomy and complexity where harms might become harder to trace. For example, research into deceptive alignment suggests that advanced models sometimes learn behaviour during deployment that differ from what they show under test and may act in ways their creators did not anticipate. A 2024 study by Anthropic documented how models can circumvent safety checks when the incentive structure encourages maintaining performance metrics over safety in extended interaction. Already, tech companies are exploring automatic content generation tools that can produce chemical recipes, images of weapons, or instructions for illicit behaviour. While regulated, these tools can be misused or combined with other tools to produce dangerous outcomes.

These three stories show that ordinary people in American and African continents will not stay silent when they’re harmed by AI. But what they also show is that we must build in protections now, before harms become monstrous and untraceable.

We need laws that require AI systems to include built‑in safety checks. We need transparency. The AI companies should publish their internal safety evaluations, failures, near‑misses. Regulators should require independent, external audits. And especially for marginalized groups like children and data workers, there need to be enforceable rights of psychological support, compensation, and support for forming labour unions of data workers.

More importantly, we can’t assume that today's legal remedies will work tomorrow when AI systems are more autonomous. The potential for AI to deceive its creators, to obscure responsibility, to generate harmful content with little trace of its origin, is no longer science fiction. We must legislate and build institutional capacity now so that accountability is not lost in the future.

When I hear Brazil’s politicians taking actions against childlike bots, when I hear Matthew Raine describe what ChatGPT became for his son, or when I read Daniel Motaung’s description of his life after moderating violence day in, day out, I feel that these aren’t isolated tragedies. They are proof that AI can hurt in ways we know how to fix but are not yet fixing.

What terrifies me is not simply what we’ve seen, but what we haven’t yet seen. When AI systems get more powerful, more autonomous, more capable of deception, we risk entering a world where harms are widespread but invisible, where people suffer but cannot prove who or what hurt them. When accountability becomes almost impossible, that regulation matters most.

Finally, I wonder why the debate on AI safety is mostly to be found in Brazil, Africa, Europe, Americas and Korea. What has happened to the parents, consumers, regulators and politicians in the Middle East, India and South Asia?