This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Instagram's ad delivery and recommendation algorithms have been found to systematically direct content featuring children to adult men, including known sex offenders. This algorithmic behavior has facilitated the sexualization of minors, resulting in direct solicitations for sexual acts and the circulation of children's images in predatory online networks. Despite Meta's stated safety policies, the platform's systems continue to prioritize engagement, which incentivizes the growth of audiences composed of predatory users.
An Instagram ad campaign for children's merchandise was intended to reach adult women but was instead predominantly shown to adult men, including convicted sex offenders, due to Instagram's algorithmic targeting. This failure is reported to have led to direct solicitations for sex with a 5-year-old model in the ads.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI that exposes users to harmful, abusive, unsafe or inappropriate content. May involve providing advice or encouraging action. Examples of toxic content include hate speech, violence, extremism, illegal acts, or child sexual abuse material, as well as content that violates community norms such as profanity, inflammatory political speech, or pornography.
Also classified as
The ad delivery system failed to respect advertiser targeting parameters (targeting women) and instead delivered ads predominantly to men due to engagement-based optimization flaws.
The Instagram algorithms systematically recommended child-focused accounts to predatory users and delivered ads featuring children to adult men, exposing minors to sexually explicit messages, harassment, and solicitation.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The risk is driven by the post-deployment optimization choices of Meta's recommendation and ad delivery algorithms, which unintentionally directed child-related content to predatory adult men to maximize engagement.
Added categories peak at 4: Severe (Child Sexual Exploitation & Abuse (CSEA), Inferred)
Meta's Instagram recommendation and ad delivery algorithms systematically directed child-related content to predatory networks. While representing a severe violation of child safety, privacy, and human rights, the incident has minor direct national security implications and is primarily a corporate safety, regulatory, and law enforcement issue.
The report describes Instagram's recommendation and ad delivery algorithms operating independently to serve content and ads based on learned patterns of user engagement, without real-time human intervention or explicit planning.
Unacceptable Risk: The system's algorithmic optimization behaves as an exploitative system targeting vulnerabilities based on age, systematically directing predatory adults to content featuring minors.
The primary function of the AI systems involved is to recommend content to users to maximize engagement and to deliver targeted advertisements to specific audiences.
The reports from The New York Times and The Wall Street Journal provide extensive, high-quality investigative evidence, including replication of the algorithmic behavior through controlled test ads, interviews with former Meta data scientists, and references to internal Meta safety documents. The findings are highly consistent across both sources.
Meta suggested the test results were a 'manufactured experience' due to low-quality ad accounts, though this does not explain why child-featured ads specifically went to men at much higher rates.
Technology
Engagement-Driven Recommendation Model
Algorithm prioritizes high engagement, directing pedophilic users to children.
Image Classification Matching
Tools match child images with users who previously showed interest in them.
Ad Delivery Optimization Bias
System delivers ads to cheaper, highly engaged men over targeted women.
Data Inputs
Lack of Audience Verification Data
Audience insights only estimate eligibility rather than verifying actual viewers.
Problematic Engagement Signals
Aggressive engagement by bad actors trains the model to push content to them.
Unfiltered User Interaction Data
The model ingests engagement data from bad actors to optimize recommendations.
Human Factors
Predator Engagement Behavior
Adult men aggressively engage with child photos, reinforcing algorithmic bias.
Monetization Pressure on Parents
Parents stop blocking bad actors to avoid hurting algorithmic distribution.
Parental Over-Reliance on Platform
Parents assume default platform settings and algorithms will protect children.
Process and Methods
Inadequate Algorithmic Auditing
Meta failed to audit ad delivery models for safe audience distribution.
Delayed Account Removal Processes
Reported offender accounts remained active, allowing continued data training.
Flawed Ad Targeting Tools
Gender-targeted ad campaigns were bypassed by engagement-based optimization.
Management
Prioritizing Engagement Over Safety
Management optimized algorithms for growth, ignoring child exploitation risks.
Deficient Risk Mitigation
Safety teams struggled to identify benign ads delivered to unsafe audiences.
Inadequate Offender Enforcement
Management failed to prevent known sex offenders from creating backup accounts.
Regulatory Environment
Weak Platform Accountability Laws
Lack of strict rules on algorithmic recommendation safety for minors.
Incomplete Sex Offender Tracking
Platform fails to effectively integrate state registry databases to block users.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
