Brand Safety Tech Firms Falsely Claimed Use of AI, Blocking Ads Using Simple Keyword Lists

Brand safety tech firms falsely claimed use of AI, blocking ads using simple keyword lists.

Risk classification

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

The brand safety systems lacked the capability and robustness to accurately analyze context, relying instead on blunt keyword matching that resulted in high error rates and misclassifications.

Additional risk subdomains

• 6.3 Economic and cultural devaluation of human effort: The systematic misclassification and subsequent defunding of high-quality journalism devalues the economic viability of professional human reporting.

Causal factors

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

The financial harm to publishers was an unintended consequence of the brand safety AI systems failing to accurately classify content post-deployment.

EU AI Act risk tier

• Risk tier: 4 Minimal or No Risk

Minimal or No Risk: The brand safety classification systems are similar to spam filters or simple content classifiers, which pose low direct risk to users and do not fall under the high-risk or prohibited categories of the EU AI Act.

AI system and alleged parties

• AI system: brand safety technology
• AI purpose: Content Moderation; Content Verification
• Behaviour type: Tool
• Alleged deployer: brand safety tech firms
• Alleged harmed parties: news sites

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. 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 Substantial, indirect Substantial
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, 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 Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Financial

Reported: Yes, the report describes significant financial losses to publishers due to blocked ads, though the total dollar amount is not explicitly quantified.

Directly caused: Legitimate publishers lost ad revenue when brand safety tools incorrectly labeled 21% to 52% of their articles as unsafe, blocking programmatic ad placements.

Indirectly caused: Ad dollars were diverted away from high-quality news sites to lower-quality or fraudulent sites that deliberately avoided blacklisted keywords.

Inferred additional harm: We infer millions of dollars in lost ad revenue for mainstream news organizations globally, given the high percentage of blocked articles (up to 52% on Vice) and the scale of programmatic advertising.

Differential treatment

Reported: Yes, the report describes how journalists covering serious topics are disproportionately affected.

Directly caused: Journalists focusing on serious topics like Middle East affairs, obituaries, or political events are disproportionately likely to have their work marked as unsafe.

Inferred additional harm: This differential treatment likely disincentivizes publishers from funding serious investigative journalism or covering sensitive but important public interest topics.

Democracy

Reported: Yes, the report describes the defunding of legitimate mainstream news organizations.

Indirectly caused: The systematic defunding of mainstream news sites (like NYT, WSJ, and Economist) weakens the financial viability of independent journalism, which is critical for democratic accountability.

Inferred additional harm: The long-term erosion of funding for high-quality journalism could lead to a less informed public and weaker democratic institutions.

People affected

• Occurrences reported: 1
• People reportedly harmed: 1000
• People reportedly exposed: 1000

Potential causes

Management

• Deceptive Marketing of AI: Vendors claim advanced AI and ML but deliver rudimentary keyword blocking.
• Prioritizing Scale Over Accuracy: Advertisers prioritize programmatic reach over ad placement quality.

Technology

• Blunt Keyword Blocking: Systems block ads based on simple keyword lists rather than semantic context.
• Lack of Image Analysis: Detection tech fails to analyze images or text inside images on webpages.
• Dynamic Content Misalignment: Crawlers miss dynamic webpage changes occurring after initial safety checks.

Data Inputs

• Inaccurate Content Labeling: Incomplete and imprecise classifiers falsely label entire domains like CBS.
• Conflicting Vendor Classifications: Different vendors disagree forty percent of the time on article safety.
• Keyword Manipulation by Fraudsters: Bad actors easily manipulate page text to bypass simplistic keyword filters.

Human Factors

• Overreliance on Automated Tools: Advertisers trust automated brand safety tech without verifying its accuracy.
• Neglecting Curated Include Lists: Marketers skip manual inclusion lists in favor of automated safety filters.

Process and Methods

• Flawed Classification Process: Categorizing entire sites based on a few pages or single words.
• Inadequate Crawler Latency: Crawlers analyze pages in milliseconds, missing post-load dynamic content.

Information quality

• Classification confidence: High
• Reason for confidence: The report provides clear, empirical data from the Adalytics study quantifying the misclassification rates (e.g., 21% of Economist, 30% of NYT/WSJ, 52% of Vice articles) and the disagreement rate between vendors (40%). The mechanism of failure (blunt keyword blocking marketed as advanced AI) is explicitly detailed.
• Ambiguities identified: The exact total financial loss in dollars is not specified, only the percentage of articles blocked.
• Alternative interpretations: None. The failure of the technology and its negative economic impact on publishers is clearly established.

Brand safety technologies marketed as advanced AI relied on blunt keyword lists, systematically misclassifying legitimate news articles as unsafe. This resulted in significant ad revenue losses for major publishers and diverted funds to lower-quality sites, presenting a minor long-term risk to the information ecosystem and the economic viability of independent journalism.

• Overall national security impact: Minor
• Scope: Multiple nations
• Primary target: No clear primary
• Other affected: United States, United Kingdom, and other global media markets
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. This represents an ongoing market and technological failure rather than an active, acute crisis requiring immediate government intervention.
• Autonomy: Human-supervised. The brand safety tools execute ad-blocking decisions automatically based on programmed rules, but operate under the parameters and oversight set by advertisers and technology vendors.
• Novelty: Evolved capability. Represents an evolution of basic keyword filtering rebranded as advanced AI, leading to widespread systemic misclassification at scale across the digital advertising industry.

Impact by dimension

• Physical security: Negligible. No physical threat, kinetic capability, or critical infrastructure compromise reported in connection with this incident.
• Information security: Minor. Flawed classification systems systematically defund legitimate news organizations while diverting ad revenue to lower-quality sites, which could indirectly degrade the quality of the broader information ecosystem.
• Sovereignty: Negligible. No direct threat to state authority, electoral integrity, or core government decision-making processes.
• Economic security: Minor. The incident caused significant, unintended financial harm to major news publishers, though it does not threaten critical national financial systems or strategic defense technologies.
• Societal stability: Minor. The systematic defunding of independent journalism poses a long-term risk to societal stability by weakening the financial viability of institutions critical for public information and democratic accountability.