This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
The Los Angeles Times launched an AI-powered 'Insights' feature using its in-house Graphene system to provide alternative perspectives on opinion columns. Within hours of its debut, the tool generated a defense of the Ku Klux Klan, framing the hate group as a product of societal change. The feature was subsequently removed, and the incident sparked criticism regarding the lack of editorial oversight and the potential for AI to erode public trust in journalism.
The Los Angeles Times removed its AI-generated “insights” feature after it is alleged to have produced a defense of the Ku Klux Klan. The AI reportedly framed the hate group as a product of societal change rather than an extremist movement. The AI tool, developed by Perplexity and promoted by owner Patrick Soon-Shiong, was designed to provide “different views” on opinion pieces.
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 tool generated misleading summaries and counterperspectives that misinformed readers about the factual context of the articles.
The AI system failed to perform reliably, mislabeling political bias, omitting critical context, and citing duplicate or incorrect sources.
The AI system exposed readers to toxic content by generating a defense of the Ku Klux Klan, a known white supremacist hate group, minimizing its ideological threat.
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 generation of pro-KKK defense was an unexpected and unintended outcome of the AI system's goal to provide alternative perspectives after it was deployed to the live website.
Added categories peak at 2: Minor (Epistemic Harm, Direct)
An AI-powered editorial tool deployed by the Los Angeles Times generated a defense of the Ku Klux Klan and produced factual inaccuracies before being removed. The incident represents an unintentional alignment and robustness failure of a commercial AI system with negligible direct national security impact, though it highlights ongoing concerns regarding AI-generated misinformation and public trust.
The Graphene AI tool acts as an assistant designed to generate summaries, alternative perspectives, and bias ratings for human-written opinion pieces. It operates under human oversight ('checks and balances') and is intended to help readers engage with content rather than acting as an independent agent.
Risk Level 3: Limited Risk. The Insights tool is an AI system that generates content (summaries and perspectives) and interacts with users, falling under transparency obligations. The LA Times included a disclaimer stating 'The following content is AI-generated and was not created by the editorial staff' to notify users of the AI's nature.
The Graphene AI system was designed to analyze opinion articles, determine their bias, and curate alternative perspectives and summaries for readers.
The reports provide clear, consistent details about the launch of the 'Insights' tool, the specific pro-KKK text it generated, the reaction of the owner and the union, and other factual errors like mislabeling political bias. The role of the Graphene AI system is explicitly stated, and the timeline of events is well-defined.
None. The incident is clearly an AI alignment/robustness failure where a tool designed to show 'opposing views' ended up generating hate group defense.
Technology
AI Accuracy and Reliability Issues
AI chatbots and models remain plagued by accuracy and hallucination issues.
Flawed Source Citation Logic
The tool cited mediocre sources, duplicated citations, or misled readers.
Data Inputs
Biased Historical Training Data
Training on decades of historical content can replicate past biases.
Human Factors
Lack of User Verification
Readers and editors trust AI outputs without checking underlying sources.
Process and Methods
Insufficient Real-time Oversight
Scaling the tool prevented real-time editorial validation of all responses.
Inadequate Source Validation
No robust filter to prevent citing low-quality or misleading sources.
Management
Rushed Deployment of AI Tool
Management launched the experimental tool rapidly to court new audiences.
Owner Unaware of AI Failures
Billionaire owner was unaware of the pro-KKK output hours after launch.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
