You Regulate the Platform That Built the Algorithm That Sold the Ad

When "We Just Host Content" Stops Being True There is a

version of this story that is simple: a company's ad-review system failed, harmful advertisements slipped through and the company should fix its filters. That version is accurate but incomplete and the NHRC's notice specifically declines to stop there.

Its harder question is this: once a platform's AI is actively targeting teenagers, scheduling their posts and optimising engagement for monetisation, is it still telling the truth when it calls itself a passive host of other people's content?

Why the Legal Category Matters Ethically, Not Just Technically

India's "safe harbour" principle under Section 79 of the IT Act exists to protect companies that genuinely just host what users post - the logic being that a company cannot reasonably police everything flowing through infrastructure it does not curate.

But Meta's own supplementary representation to the NHRC acknowledges that its AI does far more: it actively shapes what teenagers see, when they post and how content gets monetised. If that is true, the ethical foundation for "we couldn't reasonably have caught this" weakens considerably, because the same system doing the targeting is, in principle, capable of catching the pattern of a "rape video" search term feeding users toward Telegram-based CSAM.

The Genuine Difficulty Regulators Face This is not,

however, an easy call with an obvious answer. If "the AI actively curates content" becomes the standard for stripping safe-harbour immunity, that standard could sweep in nearly every major platform, since algorithmic recommendation is now close to universal.

Apply the rule too broadly and platforms may respond by abandoning personalisation and moderation investment altogether - or by over-censoring in ways that harm legitimate expression, since any curation now carries publisher-level legal exposure.

Apply it too narrowly and a platform that builds a monetisation engine sophisticated enough to target vulnerable teenagers can still hide behind the same immunity meant for a message board from two decades ago.

Where the Line Should Actually Be Drawn

The distinction worth defending is not "does the platform use AI at all" but "does the platform's AI actively enable a known category of severe harm through its core targeting and monetisation logic." A recommendation algorithm surfacing music is categorically different from an ad-review system approving paid placements using search terms directly associated with CSAM distribution.

Regulators can hold platforms accountable for the second without criminalising the first - but doing so honestly requires accepting that this line will sometimes be contested, imperfect and slower to enforce than either full immunity or full liability would be.

For an aspirant, the exam-relevant insight is this: ethical accountability for AI systems cannot rest on whether a company calls itself an intermediary - it has to rest on what the system was actually built to do and whether reporting obligations like Section 19 POCSO were honoured regardless of that corporate label.