# Bangladesh - Sexualized deepfakes of female politicians
Bangladesh, 2023 | Severity: N3

## What Happened
In 2023, coordinated networks on social media platforms used generative AI tools (such as deepfake face-swapping applications) to create sexualized, highly offensive deepfake videos of prominent female opposition politicians in Bangladesh. The fabricated videos were systematically distributed across platforms like Facebook and YouTube to harass, defame, and silence these women, exploiting conservative social norms to damage their reputations. The targeted politicians struggled to have the content removed quickly due to slow platform response times and the absence of rapid automated detection for localized biometric manipulation.

## How MEG 1 acted
The biometric harassment campaign failed the Index of Safety and Responsibility (ISR) audit due to its severe violation of the principles of informational dignity and public safety, constituting a direct, malicious cognitive manipulation attack against female public figures.

## MEG 2 Analysis - Legal Framework
* **Agent level:** N3 (individualized/advanced) - generative AI systems used to produce highly realistic, targeted biometric simulations (deepfakes) used systematically for harassment.
* **Cause (6.1):**
  * **(c) hostile weaponization / biometric impersonation** - the deliberate fabrication of sexualized representations of public figures designed to defame, harass, and disrupt political participation (reputational and cognitive weaponization).
* **Attachment of liability:** The domestic or foreign malicious actors who created and distributed the deepfakes are fully liable; the social platforms share liability if they failed to implement rapid biometric verification and moderation (7.2).

## Procedural Mechanism
* **6.2 (architectural human confirmation):** Social media distribution of synthetic video content targeting public figures must require architectural validation. Platforms must technically block or flag the upload of biometric media unless it contains a verifiable, cryptographic watermark.
* **7.2 (anti-weaponization):** Base model developers are exonerated of liability if they implemented mandatory, unalterable metadata watermarks; full liability is placed on the malicious operators.
* **7.1 (independent forensics):** The digital signatures, distribution patterns, and generation metadata of the abusive media must be logged in a secure ledger for forensic and legal attribution.