AI-generated propaganda in India targets Muslims, thrives on Instagram and X amid moderation failures

AI-generated propaganda in India is targeting Muslims - and platforms are missing it

A recent investigation uncovered a small but active network using widely available AI tools to produce and spread content that targets Muslim communities in India. Researchers reviewed over 1,300 AI-generated images and videos across nearly 300 accounts on X, Facebook, and Instagram.

The materials often use polished, visually appealing art styles that slip past moderation and hook younger viewers. The result: harmful narratives scaled at low cost, high speed, and high engagement.

What's being produced

• Conspiracy narratives: False claims that Muslims are plotting against national interests.
• Dehumanizing portrayals: Muslims depicted as threatening or less than human.
• Sexualized depictions of Muslim women: Stereotypes that reinforce bias and harassment.
• Violent scenes in popular art styles: Content that normalizes sectarian conflict while looking "artful."

Enforcement gaps and reach

Platforms largely failed to act. None of 187 flagged posts were removed.

Instagram generated close to two million interactions from this content, despite hosting fewer posts than X. This shows how style-first, high-aesthetic media can outperform text in spread and stickiness.

Why this matters for general, IT, and development teams

Unchecked AI-driven propaganda fuels social tensions, causes psychological harm, and can lead to offline violence. For teams building products, ads, or community features, the risk is brand safety, user trust, and regulatory scrutiny.

The fix isn't a single model or rule. It's layered controls, clear accountability, and transparency.

Practical steps for platforms, product, and policy teams

• Content provenance by default: Adopt cryptographic media provenance and labeling across creation, upload, and editing flows. See the C2PA standard here.
• Safety-by-design for generators: Block prompts that target protected classes with violent or sexual content. Red-team regularly against style-based evasion and release safety notes with each model update.
• Multi-modal detection: Use ensembles (vision, text, audio) plus style and behavior signals. Score for risk (protected class + violence/sexualization + coordination indicators) and prioritize human review.
• Friction and context: Apply click-through warnings, reduced distribution, and clear labels on likely synthetic or policy-risk content. Make policy rationale visible.
• Coordinated network analysis: Track small clusters that reuse prompts, styles, captions, or posting schedules. Pair this with rate limits and identity checks for repeat offenders.
• Metrics that matter: Measure time-to-detection, time-to-contain, and post-incident reach. Publish enforcement totals, restore rates, and error bounds in transparency reports.
• Researcher access: Provide safe, rate-limited APIs and datasets for external audits. Enable trusted flagger programs with SLAs during sensitive events.
• User tools: Simplify reporting flows for hate and synthesized media. Offer muted keyword/style controls and community-based fact cues.
• Localized expertise: Staff reviewers with regional language and cultural context to catch coded references and dog whistles.

What teams and individuals can do now

• Verify before you share: Treat eye-catching images and videos as unverified by default. Look for provenance labels and reverse-image search.
• Build internal guardrails: If your org publishes media, require source logs and review for protected-class harm. Keep a takedown and correction playbook ready.
• Educate your staff: Train on prompt safety, bias, and content risks. A practical starting point for teams working with generative tools is here: Prompt Engineering resources.

Policy and ecosystem moves

Stronger oversight and transparency from platforms, policymakers, and AI developers is overdue. Clear reporting, consistent enforcement, and interoperable provenance standards will reduce reach and limit harm.

For synthetic media specifically, align internal policies with industry guidance like the Partnership on AI's recommendations on responsible synthetic media.

Bottom line

A small network can use low-cost AI tools to spread high-impact hate. Without layered defenses and clear accountability, engagement wins and communities lose.

If you build or manage digital products, treat this as a reliability and safety problem. Ship safeguards with the same rigor you apply to performance and uptime.