Bug in Instagram’s “Related Hashtags” Algorithm Allegedly Caused Disproportionate Treatment of Political Hashtags

A bug was reported by Instagram’s spokesperson to have prevented an algorithm from populating related hashtags for thousands of hashtags, resulting in an allege preferential treatment for some politically partisan hashtags.

Risk classification

• Primary risk domain: 7 AI system safety, failures, & limitations
• Primary risk subdomain: 7.3 Lack of capability or robustness

The incident stems from a technical error or bug in Instagram's recommendation algorithm that caused it to fail to perform reliably and equitably across parallel search terms.

Additional risk subdomains

• 3.1 False or misleading information: The algorithm recommended and propagated hashtags containing false information and baseless smears against a presidential candidate.
• 1.2 Exposure to toxic content: Users searching for standard political terms were exposed to abusive, inappropriate, and toxic hashtags suggested by the platform.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The incident was caused by an unintentional technical bug in Instagram's deployed related hashtags recommendation algorithm.

EU AI Act risk tier

• Risk tier: 4 Minimal or No Risk

Minimal or No Risk: The AI system is a hashtag recommendation algorithm on a social media platform, which falls under general entertainment and information search applications, posing low direct safety risks under the specified categories.

AI system and alleged parties

• AI system: Instagram related hashtags algorithm (Meta)
• AI purpose: Content Recommendation; Content Search
• Behaviour type: Autonomous
• Alleged developer: Instagram
• Alleged deployer: Instagram
• Alleged harmed parties: Instagram users

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 Minor, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Minor
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Substantial
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes toxic and malicious content spread directly by the incident.

Directly caused: The algorithmic bug directly served and promoted toxic and abusive hashtags such as '#creepyjoebiden', '#joebidenpedophile', and '#joebidenisaracist' to users searching for Biden-related terms.

Inferred additional harm: It is likely that millions of Instagram users were exposed to these toxic and defamatory hashtags before the feature was disabled.

Differential treatment

Reported: The report explicitly describes differential treatment of individuals caused directly by the incident.

Directly caused: The algorithmic bug disabled related hashtags for all 10 reviewed Trump campaign terms, shielding him from negative associations, while displaying related hashtags for all 10 parallel Biden terms, steering users to insults and disinformation.

Inferred additional harm: This differential treatment likely created an unfair political advantage on the platform during the 2020 US presidential election campaign.

Democracy

Reported: The report explicitly describes the potential erosion of democratic norms caused indirectly by the incident.

Indirectly caused: The disparate treatment of presidential candidates opened the door for weaponized disinformation, potentially influencing the 2020 US presidential election and tipping the scales of equitable discourse.

Inferred additional harm: The incident likely contributed to a broader erosion of public trust in social media platforms' neutrality and the integrity of democratic election processes.

Epistemic

Reported: The report explicitly describes epistemic harm caused directly and indirectly by the incident.

Directly caused: The algorithm directly recommended and spread false and misleading conspiracy theories and baseless claims, such as '#joebidenpedophile' and '#clintonbodycount', to users.

Indirectly caused: The propagation of these false narratives gamed by bad actors undermined shared consensus reality and polluted the information ecosystem for voters seeking accurate information.

Inferred additional harm: Many voters were likely misled by these algorithmically suggested false claims, reinforcing inaccurate beliefs and political polarization.

People affected

• Occurrences reported: 1
• People reportedly harmed: 2
• People reportedly exposed: 112000000

Potential causes

Management

• Asymmetric Safety Guardrails: Management failed to ensure equal protective measures for both campaigns.
• Unpreparedness for Election: Despite safety promises, management did not prevent critical algorithmic bugs.

Technology

• Algorithmic Suggestion Bug: A technical error suppressed related hashtags for some terms but not others.
• Algorithm Vulnerability: The suggestion engine was easily gamed by bad actors to spread disinformation.

Data Inputs

• Toxic User-Generated Content: The algorithm ingested vitriolic hashtags like creepyjoebiden as inputs.

Human Factors

• Lack of Active Human Oversight: Experts noted political hashtags lacked constant human review before elections.

Process and Methods

• Ad-Hoc Policy Application: Hashtag disabling was applied inconsistently across different public figures.
• Inadequate Algorithmic Audits: The bug went unnoticed for months, showing a lack of regular system audits.

Regulatory Environment

• Unclear Campaign Finance Rules: Lax rules failed to prevent asymmetric algorithmic treatment of candidates.

Information quality

• Classification confidence: High
• Reason for confidence: The reports from Tech Transparency Project and BuzzFeed News provide clear, consistent details about the nature of the algorithmic bug, the specific hashtags affected, the platform's response, and the political context. Instagram's spokesperson confirmed the technical error, leaving little ambiguity about the occurrence of the event.
• Ambiguities identified: The exact technical cause of the bug and why it specifically affected Trump-related hashtags and not Biden-related ones remains unexplained by Instagram, beyond calling it a 'technical error'.
• Alternative interpretations: Some critics suggested the disparate treatment was intentional political bias or an in-kind campaign contribution, though Instagram maintained it was an unintentional technical bug.

An Instagram algorithmic bug in 2020 resulted in disparate treatment of US presidential candidates, shielding one from negative search terms while directing users to disinformation and smears against the other. This highlighted substantial vulnerabilities in social media recommendation systems that can be exploited or fail during critical democratic processes.

• Overall national security impact: Substantial
• Scope: Single nation
• Primary target: United States
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. The issue was resolved by disabling the feature, representing a strategic, long-term concern regarding algorithmic vulnerability in elections rather than an active crisis.
• Autonomy: Full autonomy. The recommendation algorithm generated and paired the related hashtags automatically without real-time human intervention or pre-moderation.
• Novelty: Evolved capability. While algorithmic recommendation bias is known, the specific asymmetric shielding of one candidate and targeting of another during a presidential election represents an evolved threat vector.

Impact by dimension

• Physical security: Negligible. The incident was purely digital and algorithmic, with no physical, kinetic, or critical infrastructure impact reported.
• Information security: Substantial. The algorithmic bug amplified political disinformation and negative conspiracy theories targeting a presidential candidate, impacting the information environment during a major election.
• Sovereignty: Minor. While the bug affected the information ecosystem surrounding a democratic election, it did not directly disrupt electoral infrastructure or constitutional processes.
• Economic security: Negligible. No significant economic consequences, financial system impacts, or strategic technology theft were reported in connection with this algorithmic bug.
• Societal stability: Minor. The propagation of political smears and conspiracy theories likely contributed to political polarization and epistemic harm among voters, though it did not cause active civil unrest.