Oops! They did it again: Meta’s algorithm trips over Modi’s video, but the real glitch is bigger

For a few hours in late July, one of the most-watched persons in the world vanished from two of the world's biggest platforms. Narendra Modi had put up a vertical video on July 23, his first real attempt to talk to Gen Z in its own idiom, addressing protesting students during the 36-day churn over the NEET-UG 2026 mess and promising harder laws against paper leaks. Meta's machinery took it off Facebook and Instagram. Then put it back on. The company attributed the temporary pull-out to a technical error. Nobody in GoI bought it.

The row that has followed is not really about the 5-6 hrs that the video clip stayed down. It's about a bill of grievances that has been piling up for a while. Meta's global team was called in this time. The parliamentary standing committee on IT gave the company three days to apologise, or face a review of its safe harbour, a legal provision that protects internet and social media platforms from being sued for things their users post, provided the platforms remove illegal content when told to do so.

Also Read: Mark Zuckerberg sorry for child abuse content on Meta platforms, PM video takedown


An apology did come from Meta's global affairs chief Joel Kaplan, on the company's behalf, though there are reports of Mark Zuckerberg himself conveying it. The regret, in any case, stretched past the Modi clip: to child sexual abuse material surfacing on Meta's platforms, unlabelled deepfakes and morphed videos (some aimed at Modi himself, which has now resulted in an FIR in Telangana), and to what officials politely called 'operational lapses'.

But before we reach for our pitchforks, it helps to pull 3 tangled things apart: what these algorithms actually do; what the law actually demands; and where the blame actually belongs.

The machinery Obviously, nobody at Meta sat in a room and decided to delete the Indian prime minister's post. At this scale, billions of posts a day, moderation is done by software that assigns a guilt score, and moves on. It has no idea what it is looking at.

That is precisely why the same system has, over the years, killed breast-cancer awareness posts, throttled art photographers, refused ads carrying nude paintings that have hung in galleries for ages, and quietly buried plus-size and activist accounts. Meta's own Oversight Board keeps flagging it.

The howler is boringly consistent: art, health, dark skin and any language the model reads badly. And it is gloriously non-partisan - the same 'oops' filter once buried the #ResignModi hashtag in 2021. A filter that mistakes a Renaissance nude for pornography is possibly not plotting against anyone. It is just bad at its job. Which rather deflates the charge that pulling Modi's video was a conscious editorial act by a 'publisher'.

Also Read: Meta officials meet govt over PM Modi video takedown; panel seeks Zuckerberg apology

But if that lets Meta off one hook, it hangs it on a worse one. Choosing to referee a market of a billion-plus people almost entirely through dumb automation is a decision, not a mishap. 'The algorithm did it' explains the mess, but it doesn't excuse it.

The point Meta keeps sliding past is not malice but foreseeability. A company this clever knows its filters strangle art, and mislabel the harmful. It knows deepfakes outrun its labels. It knows that wrongly nuking a head of government's account is a systemic risk, not a rounding error. Not building for the failure you can see coming is negligence, never mind the intent.

The law Which brings us to the law, and a question sharper than the shouting suggests. Information Technology Act 2000, and Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules 2021, hand out safe harbour on terms: do your diligence, run a grievance system, pull illegal content fast.

The whole idea assumes a neutral pipe carrying whatever users pour in. But a platform that ranks, recommends, boosts and, as GoI alleges, takes money to push certain content, is no pipe. It's choosing what you see. Immunity was never meant to cover that act of choosing.

So, the fix is to separate the plumbing from the editing. Let safe harbour stay strong for genuine intermediary work, and bolt on real duties wherever platforms amplify: open moderation logs, takedown records that can be audited, hard clocks for the truly illegal like child sexual abuse material (CSAM), compulsory provenance tags on synthetic media, and a proper appeal with a human at the end - the same one for a prime minister and for a protesting citizen.

The blame Where an algorithm makes the editorial call, the rule should attach to that conduct, not to the platform's mood of the moment, or of the complainant. A wider consultation would help draw the line cleanly, so that the big amplifiers cannot keep crossing it while genuinely compliant intermediaries are not throttled for someone else's sins.

Zuckerberg's reported apology should close the news cycle for now. But it settles nothing that actually matters. The video is back up now. Everything hard about scale, opacity and who gets to police the police of speech sits exactly where it did before the brouhaha.

GoI's warning was timely. But the real work - of making the rules both effective and fairly applied - has barely even begun.