Full Report
This article is the result of a collaboration with German YouTube channel STRG_F. You can watch their documentary here. Warning: This article discusses non-consensual sexually explicit content from the start. The graphic images claim to show Patrizia Schlosser, an investigative reporter from Germany. She’s depicted naked and in chains. “At first I was shocked and […] The post Faking It: Deepfake Porn Site’s Link to Tech Companies appeared first on bellingcat.
Analysis Summary
# Incident Report: Proliferation of Non-Consensual Deepfake Imagery on Hosting Platform
## Executive Summary
This summary details the discovery and investigation into the proliferation of non-consensual sexually explicit deepfake media, exemplified by an investigative journalist, Patrizia Schlosser, who discovered doctored images of herself on the platform MrDeepFakes. The incident highlights the growing threat of easily accessible AI-generated synthetic pornography, facilitated by associated AI application developers like Deepswap AI, leading to widespread data misuse and significant reputational harm for victims. Response involved investigative journalism to expose the platform and pressure on application stores to delist associated tools.
## Incident Details
- Discovery Date: Not explicitly stated; investigation resulted in public reporting "earlier this month" (relative to the article's publication). The specific compromised content mentioned (Patrizia Schlosser's album) was online for "almost two years."
- Incident Date: Ongoing proliferation since the inception of the MrDeepFakes platform (active since at least February 2018).
- Affected Organization: Not a single organizational breach, but multiple individuals (journalists, actresses, musicians, politicians, and non-public figures) as victims.
- Sector: Media/Journalism, Technology, Entertainment.
- Geography: Global operation (platform hosting bounced globally, financial activities obscured; associated apps traced to Hong Kong, Malta, and global app stores).
## Timeline of Events
### Initial Access
- Date/Time: Ongoing since platform inception (February 2018).
- Vector: Upload/submission of source material (a single photo of a target) by users to platforms like MrDeepFakes, potentially facilitated by accessible creation tools or services.
- Details: Attackers exploit AI technology (deepfakes) to generate non-consensual explicit content featuring any individual whose photo is available.
### Lateral Movement
- *Not applicable in a traditional sense; this is a platform exploitation/content hosting incident rather than a network intrusion.* The content spreads via user access to the MrDeepFakes website and potentially associated promotional applications.
### Data Exfiltration/Impact
- Content hosted on MrDeepFakes had been viewed almost two billion times.
- Victims, such as Patrizia Schlosser, suffered significant emotional distress/reputational damage from non-consensual images appearing online for extended periods (her album lasted almost two years).
### Detection & Response
- Detection: Through investigation by investigative journalists (Patrizia Schlosser and STRG\_F/Bellingcat collaboration).
- Response actions taken: Investigative partners tracked down the source of the content, leading to its eventual removal from the specific profile/album. Investigative pressure led to the suspension of associated apps from major stores (e.g., Google suspended DeepSwap PRO).
## Attack Methodology
- Initial Access: Acquisition of a base photo of the target, leveraged by users of the MrDeepFakes platform or affiliated AI tools.
- Persistence: Content remains accessible on the MrDeepFakes platform, which utilizes global hosting bounces and cryptocurrency payments to obscure ownership and evade takedowns.
- Privilege Escalation: *Not applicable.*
- Defense Evasion: Obscuring financial ownership via cryptocurrency transactions (new address per transaction) and bouncing hosting providers globally.
- Credential Access: *Not applicable.*
- Discovery: Users discover existing content or create new content using the advertised AI tools.
- Lateral Movement: *Not applicable.*
- Collection: Users upload base photos to creation apps or directly to the centralized MrDeepFakes repository.
- Exfiltration: Content is made available for viewing/download by the site's membership base (nearly 650,000 members).
- Impact: Severe reputational damage, emotional and psychological harm to victims due to the public and widespread nature of the sexual content.
## Impact Assessment
- Financial: Not explicitly calculated, but the monetization method involves premium memberships via cryptocurrency. Associated application developers (e.g., Deep Creation Limited) appear to be internationally registered businesses.
- Data Breach: Non-consensual sexually explicit images and videos created using victim likenesses.
- Operational: No disclosed organizational downtime, but significant personal operational disruption for targeted individuals (e.g., Patrizia Schlosser actively engaged in counter-investigation).
- Reputational: Severe reputational damage to victims, as content is viewed billions of times globally.
## Indicators of Compromise
- Network indicators: The specific platform URL (MrDeepFakes) and linked promotional URLs/apps (Deepswap AI). *(URLs/IPs defanged due to policy)*.
- File indicators: Non-consensual sexually explicit deepfake image files hosted on the platform.
- Behavioral indicators: User transactions utilizing cryptocurrency to purchase premium access on the platform.
## Response Actions
- Containment measures: Investigative partners pressed platform stores (Google Play, Apple Store) leading to the suspension/removal of associated creation applications (e.g., DeepSwap PRO).
- Eradication steps: Targeted removal of specific content albums (e.g., Schlosser’s album removed).
- Recovery actions: Victims engaged in counter-narrative and legal tracking efforts, sometimes facilitated by press investigation (STRG\_F/Bellingcat).
## Lessons Learned
- The accessibility of robust AI tools makes it trivial to manufacture high-impact synthetic abuse material using only a single source photo, affecting non-public figures as easily as celebrities.
- The operational structure of deepfake platforms (global hosting, crypto finance) is highly effective at evading immediate takedown and attribution attempts.
- Public pressure resulting from collaborative investigative journalism can force application stores to remove associated creation tools, demonstrating a temporary mitigation vector.
## Recommendations
- Advocate for stronger, harmonized international legislation specifically criminalizing the creation and sharing of non-consensual intimate images, including deepfakes.
- Technology platforms (app stores, hosting providers) must establish robust, rapid-response mechanisms to detect and suspend services demonstrably facilitating illegal synthetic pornography distribution.
- Users of AI creation tools should be monitored for patterns indicative of illicit content generation, and application developers must maintain better transparency regarding their operational locations and ownership structures to facilitate legal accountability.