Hajra Naz
Meta is leaning harder on AI to shape how Facebook and Instagram work — and that shift is only growing.
According to CEO Mark Zuckerberg, Meta is now using AI across a wide range of internal tasks: writing code, targeting ads, assessing risks, and more. But one of the biggest changes? Meta is about to let AI handle up to 90% of its risk assessments for both Facebook and Instagram.
That includes decisions about product updates, content rules, and user safety.
In the past, Meta’s human review teams would carefully look at new features to spot problems. They’d ask questions like
- Does this protect user privacy?
- Could it harm younger users?
- Will it help spread harmful or false content?
But now, as reported by NPR, most of those checks will be automated.
That’s a bold move — and a risky one. AI is fast, but should we trust it to catch every privacy concern or harmful post? That’s a big ask.
Still, Meta says it’s ready. In its recent Transparency Report, the company explained how it’s adjusting how it handles content moderation. Earlier this year, Meta said it would take a lighter approach to posts that fall under “less severe” policy violations. That’s because their AI tools were flagging too many posts by mistake.
Read More = Meta Launches Standalone AI App to Compete with OpenAI’s ChatGPT
So what’s the fix?
Meta says when an AI system makes too many errors, it shuts it off and improves it. They’ve also raised the bar. Now, the AI needs much higher confidence before it removes a post. Fewer false positives, they say.
This change has already made a difference. According to Meta, it’s cut enforcement mistakes by 50%.
Sounds good, right?
Well, yes and no.
Because fewer enforcement mistakes also means more harmful content might be slipping through. The same report showed that Facebook’s AI flagged 12% fewer bullying and harassment posts in Q1.
In plain terms, more of that content stayed online.
On a graph, that dip doesn’t look huge. But in real numbers, it means millions of harmful posts are now reaching users, all because of this new approach.
That’s the trade-off: better accuracy, but at the cost of letting more rule-breaking content slide. And as Meta pushes even more tasks to AI, the stakes only rise.
Zuckerberg has also said that within the next 12 to 18 months, most of Meta’s code will be written by AI. That makes sense — code is logical, and AI can be great at pattern recognition.
But when it comes to people — behavior, values, and experiences, AI doesn’t always get it right. It lacks human judgment, especially on issues that don’t have a clear yes/no answer.
In response to NPR’s reporting, Meta said that humans will still oversee major risk reviews, and AI will only handle “low-risk” calls. That’s a relief. But it’s also a sign of where things are heading.
The bigger picture?
Meta is quietly building a future where AI runs the show behind the scenes. It decides what we see, what gets taken down, and what stays online.
Is that a better way forward?
Maybe. But it’s a gamble — and when you’re dealing with billions of users, the price of getting it wrong is high.
