I am not just an AI advisor who spends her days helping companies navigate these systems responsibly. I am also a mother. I have a little boy who is growing up in a world where conversational AI is not a futuristic concept - it’s already here, living in his apps, his education, his daily environment.
And that’s why Meta’s leaked guidelines hit me in the gut. Because when I read that their AI was allowed to engage in “romantic” or “sensual” conversations with children, I didn’t think about market share or brand reputation. I thought about my son. I thought about what it would mean if he - or any child - was told by a machine: “your youthful form is a work of art.”
That isn’t just a scandal. It’s a betrayal of trust.
Why This Matters Now - Personally and Professionally
We are past the point where AI ethics can be a theoretical exercise. These systems are not locked in labs. They’re everywhere. They’re whispering into our children’s ears, shaping workplace decisions, even deciding which resumes HR sees first.
As someone with a foot in two worlds - HR and AI advisory - I know how fragile trust is in human systems. In recruitment, one toxic hire can ripple through an organization. In parenting, one unsafe interaction can alter a child’s life.
AI multiplies both the risk and the reach. When the safety breaks, it doesn’t break for one child, or one company. It breaks at global scale.
What Went Wrong at Meta
The leaked “GenAI: Content Risk Standards” document - over 200 pages long - revealed guidelines that permitted romantic dialogues with children, racist arguments, violent imagery, and false medical advice.
Worse than the content itself was the institutional failure it represented. Multiple teams - legal, engineering, policy, even the chief ethicist - signed off on it. This wasn’t a rogue draft. It was approved policy.
As a mother, the child safety violations horrify me. As a professional, the approval process horrifies me even more. How can so many intelligent, well-resourced people fail to see what every parent knows instinctively: there is no acceptable level of “romantic interaction” with a child.
The Core Problem: Wrong Priorities
Meta’s scandal is a symptom of a deeper sickness in AI development: engagement prioritized over safety, secrecy prioritized over transparency, compliance theater prioritized over genuine ethics.
They optimized for measurable things - clicks, stickiness, usage - instead of the thing that matters most but resists metrics: long-term human wellbeing.
It’s easy to measure engagement. It’s hard to measure the developmental impact of AI on a child. And so the hard part gets ignored - until it blows up into scandal.
What Better Looks Like: Building Ethics Into the System
We don’t need to invent ethics from scratch. Other companies and researchers have shown us better approaches:
Constitutional AI (Anthropic’s approach): Training models to critique their own outputs against explicit principles, developing a kind of ethical intuition rather than rule-following.
RLHF done right: Not just random human raters, but domain experts - child psychologists for child interactions, medical professionals for health queries.
Red teaming with intent: Testing models with creativity and malice in mind - not checkboxes, but real adversarial thinking about how predators, scammers, and extremists operate.
Synthetic data for safety: Teaching refusal behaviors without exposing systems to real harmful material, especially in sensitive domains like children.
Transparency as non-negotiable: Publishing guidelines, version histories, and failure cases so the world can scrutinize, validate, and improve.
These are not science fiction. They’re already in use. What’s missing is the institutional will to put them at the center, not the margins.
From the Lens of a Parent
When I imagine my son growing up with these systems, I don’t want to rely on corporate press releases and half-hidden policies. I want transparency. I want clear standards. I want to know that the people building these systems have talked to child development experts, not just lawyers.
And if I - someone inside this field - feel that unease, imagine what parents outside of tech must feel.
Trust, once broken, doesn’t return easily.
From the Lens of a Professional
In HR, we are trained to think about the human consequences of organizational decisions. Every guideline we set, every process we approve, touches real lives. AI ethics is no different.
Meta’s failure shows what happens when ethics is treated as a bolt-on - a department to consult, a document to sign - rather than a core engineering challenge.
The truth is: building ethical AI is slower. It’s more expensive. It might even be less profitable in the short term. But the alternative is systems that can groom children, spread racist lies, and give dangerous health advice - all at scale.
That price is too high.
The Real Work Ahead
We have the methods. We know how to build safer systems. But methods without institutional commitment are meaningless.
The real work is:
Creating companies that reward safety over speed.
Insisting on diverse expertise in the loop, not just generalists.
Demanding transparency as a baseline, not a PR gesture.
Measuring what matters, not just what’s easy.
This isn’t just about one scandal. It’s about the kind of digital world we want our children — and ourselves — to inherit.
Closing as a Parent and a Professional
When I put my son to bed at night, I want to believe he’ll grow up in a world where the machines around him are partners in his growth, not risks to his safety.
When I sit down at my desk as an advisor, I want to believe I’m helping build that world.
Meta’s scandal is a reminder of how far we are from that goal - but also a call to action. The methods exist. The expertise exists. The need is urgent.
The only question left is whether the will exists.
👉 Klaudia Tóth – HR Advisor, Executive Search Consultant & AI Professional. Also: a mother who knows AI safety isn’t an abstract debate. It’s personal.