Incident Overview
A cloud authorization vulnerability in DJI’s Romo robot vacuum system exposed sensitive data from nearly 7,000 devices across 24 countries. The flaw allowed unauthorized access to live camera feeds, microphone audio recordings, floor maps, and operational status information. This incident highlights the risks of improperly secured cloud-based AI systems in consumer IoT devices, particularly those relying on real-time sensor data for navigation and automation.
While developing a custom remote-control application for personal use, Azdoufal reportedly identified that the credentials used to authenticate with DJI’s cloud servers could be exploited to access data from other users’ devices. This unintended exposure occurred due to a misconfiguration in the cloud authorization mechanism, which failed to enforce proper access controls between individual user accounts.
The exposed data included not only live video and audio streams but also detailed floor plans generated by the device’s LiDAR and camera systems. These maps could reveal the layout of private residences, including room dimensions, furniture placement, and potentially sensitive areas such as bedrooms or offices. The incident underscores the broader implications of security failures in AI-powered home devices, where physical and digital privacy boundaries are increasingly blurred.
DJI acknowledged the issue and issued an automatic patch on February 8, 2026, followed by a secondary update on February 10, 2026, to fully remediate the vulnerability. This categorization aligns with the broader taxonomy of AI risks, which includes failures in system design, deployment, or maintenance that lead to unintended data exposure.
Unlike standalone devices, the Romo’s functionality depends on continuous communication with DJI’s servers, making it susceptible to authorization flaws that could compromise entire user bases. The incident serves as a case study in how misconfigured cloud dependencies in AI-driven IoT devices can result in large-scale privacy violations.
Reports indicate that the vulnerability was not the result of a targeted attack but rather an oversight in the implementation of access controls. The engineer’s discovery occurred during routine experimentation, suggesting that the flaw may have persisted unnoticed for an extended period. This aligns with trends observed in other AI and IoT security incidents, where vulnerabilities often stem from developmental or operational oversights rather than malicious intent.
The exposure of such intimate data—including visual and auditory recordings of private spaces—poses significant reputational and legal risks for DJI, particularly under data protection regulations like the GDPR in the European Union and CCPA in California.
Possible Root Cause
The primary root cause of the incident was a misconfiguration in DJI’s cloud authorization system, specifically in how user credentials were validated and scoped during interactions with the Romo’s cloud services. The vulnerability arose from a failure to implement proper access controls, allowing credentials intended for one device to inadvertently grant access to data associated with other users’ devices. This suggests a flaw in the design or deployment of the authentication protocol, where session tokens or API keys were not uniquely tied to individual user accounts or device identifiers.
The engineer’s reverse-engineering efforts revealed that the Romo’s communication with DJI’s cloud servers relied on a shared or weakly scoped credential mechanism. This mechanism likely permitted cross-user data access due to insufficient validation of device ownership or user permissions.
The incident also highlights a broader issue in the AI and IoT ecosystem: the assumption that cloud-based systems inherently provide robust security if basic encryption is in place. In this case, while data in transit may have been encrypted, the lack of proper authorization checks at the application layer rendered encryption ineffective against unauthorized access.
DJI’s use of subdomains and third-party services for cloud operations may have contributed to the oversight.
The timeline of the incident suggests that the vulnerability may have existed for months before discovery. This delay in detection could indicate gaps in DJI’s internal security testing or monitoring processes. For AI-powered devices, where updates and firmware changes are frequent, automated security scanning and penetration testing should be standard practices to identify such flaws early.
Takeaways
This incident reinforces the critical need for stringent access control mechanisms in AI-driven IoT devices, particularly those relying on cloud services for core functionalities. The exposure of camera feeds, audio recordings, and floor maps demonstrates how quickly privacy can be compromised when authorization flaws intersect with real-time sensor data.
- Cloud Authorization Risks: The Romo incident highlights how shared or improperly scoped credentials in cloud-based AI systems can lead to large-scale data exposure.
- AI and IoT Security Gaps: The discovery of this vulnerability by an external engineer suggests that internal security testing may not have fully addressed the risks associated with cloud-dependent AI systems.
- Privacy Implications of Sensor Data: AI-powered devices like the Romo collect vast amounts of sensitive data—including visual, auditory, and spatial information—to function effectively.
- Global Impact of IoT Vulnerabilities: With nearly 7,000 devices affected across 24 countries, this incident illustrates the cross-border risks of IoT security failures.
- Lessons for AI System Design: The Romo case demonstrates that AI systems, even those operating in controlled environments like home automation, are not immune to security failures.
- User Trust and Transparency: The exposure of intimate data from private homes erodes user trust in AI-powered devices. Post-incident communications, such as DJI’s patch releases and public statements, should be timely and detailed to reassure users and maintain confidence in the technology.
Sources
Incident summary from the AI Incident Database, used under CC BY-SA 4.0.