Defamation via AutoComplete

Google's autocomplete feature alongside its image search results resulted in the defamation of people and businesses.

Risk classification

• Primary risk domain: 3 Misinformation
• Primary risk subdomain: 3.1 False or misleading information

The autocomplete algorithm generated false and misleading associations linking innocent individuals and businesses to crimes, fraud, prostitution, and bankruptcy, causing severe reputational and material harm.

Additional risk subdomains

• 1.1 Unfair discrimination and misrepresentation: The French lawsuit alleged that autocomplete systematically associated the term 'Jewish' with prominent figures, creating an illegal ethnic file.
• 7.3 Lack of capability or robustness: The search algorithms lacked the robustness to filter out defamatory or false associations with personal names.

Causal factors

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

The defamatory associations were generated automatically by Google's post-deployment search algorithms as an unexpected outcome of predicting user queries.

EU AI Act risk tier

• Risk tier: 4 Minimal or No Risk

Minimal or No Risk: The autocomplete feature is a search utility that poses low inherent risk, similar to simple applications like spam filters, though its outputs in these specific cases caused reputational harm.

AI system and alleged parties

• AI system: Google Autocomplete (Google)
• AI purpose: Predictive Typing; Content Search
• Behaviour type: Assistant
• Alleged developer: Google
• Alleged deployer: Google
• Alleged harmed parties: Varied

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

Financial

Reported: The reports describe financial losses including job loss, business disruption, and court-ordered damages against Google.

Directly caused: Google was ordered to pay damages and costs, including 200,000 AUD to Trkulja, 300,000 yen to the Japanese plaintiff, and 3,800 euros to the Italian plaintiff.

Indirectly caused: The Japanese plaintiff lost his job and was repeatedly rejected for new work, and businesses like the Ballymascanlon Hotel and businessman AB faced potential loss of clients.

Inferred additional harm: Widespread unquantified financial losses are likely for other individuals and businesses whose reputations were damaged by false autocomplete suggestions.

Malicious content

Reported: The reports describe the generation and spread of defamatory and toxic terms associated with innocent individuals.

Directly caused: The autocomplete algorithm directly generated and displayed terms like 'hit man', 'criminal', 'prostitute', 'escort', and 'fraudster' alongside personal names.

Inferred additional harm: Widespread automated generation of toxic or defamatory search suggestions for other individuals globally.

Differential treatment

Reported: The reports describe potential differential treatment through the systematic association of ethnic terms with public figures.

Directly caused: The French lawsuit alleged that Google's autocomplete systematically associated the term 'Jewish' with certain prominent figures, creating an illegal ethnic file.

Inferred additional harm: Potential algorithmic bias or stereotyping of individuals based on automatically generated ethnic or religious associations.

Civil rights

Reported: The reports describe violations of individual personality and reputational rights.

Directly caused: Courts in Germany and Japan ruled that the automated suggestions directly violated the plaintiffs' 'right of personality' and privacy.

Inferred additional harm: Widespread unaddressed violations of individual personality and privacy rights globally due to automated search suggestions.

Psychological

Reported: The reports describe emotional distress, anxiety, and fear of reputational damage among the affected individuals and their families.

Directly caused: The automated suggestions directly caused significant distress and anxiety to victims, such as brides 'in tears' over the hotel receivership rumor and Trkulja's fear for his family's reputation.

Inferred additional harm: It is highly likely that many other unnamed individuals globally have experienced similar psychological distress due to automated defamatory suggestions.

Epistemic

Reported: The reports describe the generation of false associations that misled users and damaged reputations.

Directly caused: The autocomplete algorithm directly generated false search predictions (e.g., 'Ballymascanlon Hotel receivership') that misled users into believing false facts.

Indirectly caused: Users made real-world decisions, such as denying employment or panicking over wedding plans, based on these false beliefs.

Inferred additional harm: Widespread pollution of the information ecosystem regarding individual reputations due to automated search suggestions.

People affected

• Occurrences reported: 7
• People reportedly harmed: 8
• People reportedly exposed: 1000000

Potential causes

Management

• Prioritizing Automation Over Safety: Management relied on automated systems to avoid manual content auditing.
• Slow Response to Legal Notices: Google resisted taking down defamatory content even after direct warnings.
• Defensive Legal Strategy: Management chose to litigate rather than proactively fix systemic issues.

Technology

• Flawed Autocomplete Algorithm: Algorithmic predictions prioritized popularity over factual accuracy.
• Lack of Disambiguation: Algorithms failed to distinguish between different people with the same name.
• Lack of Defamation Safeguards: Systems lacked automated filters to block defamatory terms for individuals.

Data Inputs

• Biased Historical Search Data: Predictions relied on previous search trends and rumors typed by users.
• Web Indexing of Unverified Content: Unverified third-party content was scraped and indexed by search crawlers.
• Name Collisions in Data: Generic names caused data from different individuals to be merged.

Human Factors

• User Search Behavior: Users actively searched rumors, reinforcing defamatory autocomplete terms.
• User Cognitive Bias: Users assumed adjacent photos or terms implied a criminal connection.
• Lack of User Verification: Users accepted automated suggestions without verifying their truthfulness.

Process and Methods

• Inadequate Notice-and-Takedown: Processes for removing defamatory terms after notification were slow.
• Inconsistent Filtering Policies: Filters existed for piracy but not for personal defamation or libel.
• Lack of Pre-emptive Monitoring: No process existed to monitor predictions for potential harm before release.

Regulatory Environment

• Vague Intermediary Liability Laws: Conflicting global laws left search engine liability open to interpretation.
• Lack of Right to be Forgotten: Absence of uniform deletion rights allowed defamatory links to persist.
• US Safe Harbor Exemptions: Rigorous actual malice standards shielded platforms from negligence claims.

Information quality

• Classification confidence: High
• Reason for confidence: The reports are from highly reputable sources detailing multiple court cases across different jurisdictions with consistent facts about how Google's autocomplete algorithm works and the specific harms caused to individuals.
• Ambiguities identified: None of significance; the technical mechanism of autocomplete is clearly explained and agreed upon by both Google and the courts.
• Alternative interpretations: Google argued it was a neutral host merely reflecting user behavior rather than a publisher of defamatory content, which is a legal interpretation difference rather than a factual ambiguity.

A series of international legal cases highlighted how Google's automated autocomplete and image search algorithms unintentionally generated defamatory and false associations for individuals and businesses globally. While causing notable personal, reputational, and financial harm, the incidents represent localized civil and regulatory challenges rather than systemic threats to national security.

• Overall national security impact: Minor
• Scope: Multiple nations
• Primary target: No clear primary
• Other affected: Australia, Germany, Italy, Japan, France, Ireland, Hong Kong
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. Represents an ongoing, systemic challenge regarding algorithmic content moderation and liability rather than an immediate national security crisis.
• Autonomy: Full autonomy. The autocomplete algorithm generated and displayed search predictions dynamically and automatically without real-time human oversight or validation.
• Novelty: Evolved capability. Represented a significant evolution in how automated search indexation and predictive text algorithms can generate novel legal liabilities and reputational harms.

Impact by dimension

• Physical security: Negligible. The incident involves search engine autocomplete suggestions causing reputational and financial harm, with no impact on physical security or critical infrastructure.
• Information security: Minor. Unintentional algorithmic generation of false associations affected public figures and institutions, but lacked coordinated state-sponsored information warfare characteristics.
• Sovereignty: Negligible. While a former first lady of Germany was affected, the incident did not threaten state authority, electoral integrity, or core government operations.
• Economic security: Negligible. Financial impacts were localized to individual defamation damages and minor business disruptions, posing no threat to national economic or technological security.
• Societal stability: Minor. The algorithm caused minor societal impact through automated propagation of defamatory claims and potential ethnic categorization, violating individual privacy and personality rights.