Meta is replacing its human content moderators with large language models (LLMs) to cut costs. The shift raises a sharp question for India, its largest market: can automated systems police abuse across 22 official languages when they struggle with anything other than English?
Meta has already swapped about 50% of its human review requests for LLMs this year, and that could cross 90% for some content types by the end of 2026, The420 reported, citing the Financial Times. The company spent roughly $5 billion on content moderation in 2025, most of it on third-party contractors, and automating the work frees up that capital for Chief Executive Mark Zuckerberg’s artificial intelligence (AI) push.
What Meta says: In its Q3 2025 Integrity Report, Meta said it will transition to enhanced AI models over the next few years, “requiring less human review in areas that can be easily automated” while “focusing human expertise on the most sensitive and difficult areas of enforcement.” It said the newer models reduce mistakes, citing a test in which LLMs outperformed both its existing automation and human reviewers at detecting celebrity impersonation scams.
The India stakes are large: India is Meta’s single biggest market, with the highest number of Facebook, Instagram, and WhatsApp users of any country. Moderation here has long been strained by the country’s linguistic diversity, and handing routine enforcement to AI tests the systems where they are weakest.
Why AI moderation struggles in non-English languages: Multilingual models perform far better in high-resource languages such as English than in medium- and low-resource ones, a gap researchers trace to disparities in the availability of training data, according to a Cambridge Forum on AI paper on Meta’s moderation in the Global South. Three failure modes stand out:
- Toxicity goes undetected: Google’s Perspective API underestimates toxicity in Hindi and Swahili while rating similar English content more accurately.
- Translation strips nuance: The systems lean on machine-translated text that loses the slang, idioms, and cultural context that determine whether a post is abusive.
- Code-mixing breaks them: Indians rarely post in one clean language. In one test, adding a single Yoruba word to an English prompt produced inaccurate outputs.
The risk runs both ways: harmless posts get removed, and genuinely harmful content slips through.
Does any law require a human in the loop? No jurisdiction mandates human moderation outright. The European Union’s Digital Services Act (DSA) builds human review into its appeals safeguard: Article 20(6) says platform complaint systems cannot run “solely on the basis of automated means.” India’s Information Technology (IT) Rules, 2021, require intermediaries to run grievance redressal mechanisms.
They stop short of the DSA’s safeguard: an automated moderation decision carries no guaranteed route to human review. As…
Source link
Disclaimer
We strive to uphold the highest ethical standards in all of our reporting and coverage. We blogs.grocliq.com want to be transparent with our readers about any potential conflicts of interest that may arise in our work. It’s possible that some of the investors we feature may have connections to other businesses, including competitors or companies we write about. However, we want to assure our readers that this will not have any impact on the integrity or impartiality of our reporting. We are committed to delivering accurate, unbiased news and information to our audience, and we will continue to uphold our ethics and principles in all of our work. Thank you for your trust and support.
Website Upgradation is going on for any glitch kindly connect at [email protected]
