Images Captured by iRobot's Roomba Containing Device Users Posted on Private Online Groups

Images which were collected in an R&D project with user consent by iRobot's Roomba J7 robot vacuum showing device users sometimes in private settings were shared on closed social media groups by Venezuelan gig workers who labeled items in the images, breaching data agreements.

Risk classification

• Primary risk domain: 2 Privacy & Security
• Primary risk subdomain: 2.1 Compromise of privacy by obtaining, leaking or correctly inferring sensitive information

The incident involved the unauthorized sharing and leakage of highly sensitive, non-public personal images captured inside private homes by development devices.

Causal factors

• Entity: Human
• Intent: Unintentional
• Timing: Pre-deployment

The privacy leak was caused by the actions of human contract workers who shared private images on social media, and the incident occurred during the pre-deployment training phase of the AI model.

EU AI Act risk tier

• Risk tier: 4 Minimal or No Risk

Minimal or No Risk: The Roomba J7 is a consumer smart home appliance, which generally poses low or negligible risk to society under the EU AI Act, though its data collection practices raise significant privacy concerns.

AI system and alleged parties

• AI system: Roomba J7, Scale AI (Scale AI, iRobot)
• AI purpose: Smart Devices; Navigation Assistant
• Behaviour type: Autonomous
• Alleged developer: iRobot
• Alleged deployer: iRobot
• Alleged harmed parties: Scale AI, Roomba J7 device owners in project IO, iRobot

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Minor, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Minor, indirect Negligible
• Psychological: direct Minor, indirect Negligible
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes sensitive and inappropriate private images being shared on social media platforms.

Directly caused: Private, intimate images of individuals (including a minor and a woman in a compromising position) were shared by contract workers on social media platforms like Facebook and Discord.

Civil rights

Reported: The report explicitly describes violations of the fundamental human right to privacy.

Directly caused: The incident directly violated the privacy rights of the individuals depicted in the leaked photos, including a minor, whose faces and intimate moments were exposed without proper consent or safeguarding.

Inferred additional harm: It is likely that many other participants in the data collection program had their privacy rights compromised due to systemic lack of data protection in the annotation supply chain.

Privacy

Reported: The report explicitly describes severe privacy violations where intimate photos of individuals were leaked.

Directly caused: Personally identifiable information, including the faces of a young boy and a woman, along with intimate household scenes, were leaked to closed social media groups by Scale AI contract workers.

Inferred additional harm: Given that iRobot shared over 2 million images with Scale AI, it is highly likely that numerous other individuals had their private data exposed to contract workers without robust access controls, leading to further unquantified privacy violations.

Psychological

Reported: The report explicitly describes the highly intimate and compromising nature of the leaked photos, which would naturally cause severe distress and embarrassment to the affected individuals.

Directly caused: The young woman and the young boy depicted in the leaked photos suffered severe privacy violations and potential emotional distress due to their private moments being shared on social media.

Inferred additional harm: It is highly likely that other individuals whose private household photos were shared in these closed groups experienced significant distress, anxiety, and embarrassment upon learning of the leak.

People affected

• Occurrences reported: 1
• People reportedly harmed: 2
• People reportedly exposed: 15

Potential causes

Management

• Functionality Prioritized Over Privacy: Focus was on shipping functional robots rather than securing data pipelines.
• Inadequate Vendor Oversight: Failure to police and control how third-party annotators managed raw images.

Technology

• Lack of On-Device Anonymization: No automatic blurring of faces or sensitive objects before cloud upload.
• Invasive Hardware Sensors: High-definition cameras captured intimate household scenes from low angles.

Data Inputs

• Raw Real-World Image Collection: Over 95 percent of training data sourced from actual, unedited home spaces.
• Inclusion of Biometric Data: Faces and minors were captured and not stripped from the training datasets.

Human Factors

• Worker Sharing on Social Media: Contractors posted screenshots to Facebook and Discord to get labeling help.
• Tester Failure to Clear Space: Paid collectors did not remove children or sensitive scenes from rooms.

Process and Methods

• Use of Crowdsourced Platforms: Relying on distributed global gig workers instead of secure, in-house staff.
• Inadequate Filtering Systems: Failing to automatically detect and delete compromising images from servers.

Regulatory Environment

• Weak Federal Privacy Laws: Lack of comprehensive US laws allowed broad use of data for improvement.
• Low Legal Barriers for Consent: Vague privacy policies allowed companies broad discretion to share data.

Information quality

• Classification confidence: High
• Reason for confidence: The report from MIT Technology Review is highly detailed, featuring direct confirmations from iRobot and Scale AI regarding the source of the images, the nature of the data collection program, and the breach of NDAs by contract workers. The facts of the leak, the parties involved, and the systemic issues in the data annotation pipeline are clearly established.
• Ambiguities identified: The exact number of individuals whose photos were leaked in total remains slightly unclear, as only 15 screenshots were obtained by the journalists.
• Alternative interpretations: None. The incident is clearly a data privacy breach originating from an AI training data pipeline.

A data breach involving the leak of intimate household images captured by development versions of iRobot Roomba vacuums. While the incident represents a severe privacy violation for affected individuals and highlights systemic vulnerabilities in the global AI data annotation supply chain, its direct national security impact is minor.

• Overall national security impact: Minor
• Scope: Multiple nations
• Primary target: No clear primary
• Other affected: United States, Japan, France, Germany, Spain, Venezuela
• Alleged perpetrator: Scale AI contract workers

Threat characteristics

• Imminence: Long-term. Represents an historical incident reflecting systemic data supply chain risks rather than an active, immediate crisis.
• Autonomy: Human-supervised. The robot vacuum operated autonomously to capture images, but the data leak was executed by human contract workers in the data supply chain.
• Novelty: Evolved capability. Represents an evolved threat where autonomous consumer devices feed data into global AI training pipelines, creating new vectors for privacy leaks.

Impact by dimension

• Physical security: Negligible. The incident involved consumer smart home devices leaking personal images, with no impact on physical security or critical national infrastructure.
• Information security: Negligible. The leaked data consisted of consumer household images, with no compromise of classified intelligence, military systems, or state-level information warfare.
• Sovereignty: Negligible. The incident only affected private consumer devices and did not impact state authority, electoral systems, or government decision-making.
• Economic security: Minor. Highlights supply chain vulnerabilities in AI data annotation involving global gig workers, affecting corporate relationships, but did not threaten strategic national economic security.
• Societal stability: Minor. The incident represents a severe violation of individual privacy rights for the affected users, but did not scale to a threat to broader national societal stability or civil liberties.