Ars Technica Retracted Article After Purportedly AI-Generated Text Was Presented as Direct Quotes From Matplotlib Maintainer

Ars Technica retracted an article after purportedly AI-generated text was presented as direct quotations from a source who disputed having said them. The editor-in-chief reportedly acknowledged a standards failure and said the publication was not consistent with Ars policy on AI-generated material, and apologized to readers and to matplotlib maintainer Scott Shambaugh. An author reportedly later said he had inadvertently used an AI-paraphrased version of the source text as quotes.

Risk classification

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

The AI system generated false quotes attributed to a real person, which were then published in a news article, spreading misinformation to the public.

Additional risk subdomains

• 5.1 Overreliance and unsafe use: The reporter overrelied on ChatGPT to extract source material without verifying the output, leading to the publication of fabricated quotes.
• 6.2 Increased inequality and decline in employment quality: The incident led to the termination of the reporter and highlights the pressure on journalists to integrate AI tools, degrading working conditions.

Causal factors

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

The incident was caused by ChatGPT generating fabricated quotes, which was an unintentional hallucination occurring after the model's deployment.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The system involved is ChatGPT, a chatbot used to generate text, which falls under the category of AI systems requiring transparency about their AI nature.

AI system and alleged parties

• AI system: ChatGPT (OpenAI)
• AI purpose: Writing Assistant; Technical Text Generation
• Behaviour type: Assistant
• Alleged developer: OpenAI
• Alleged deployer: Benj Edwards, Ars Technica
• Alleged harmed parties: Scott Shambaugh, Open-source maintainers, Epistemic integrity, Ars Technica readers

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

Epistemic

Reported: The report explicitly describes epistemic harm through the generation and publication of fabricated quotes attributed to a real person.

Directly caused: ChatGPT generated fabricated quotes that were published in an Ars Technica article, misleading readers who read the piece before its retraction.

Indirectly caused: The incident contributed to the erosion of public trust in journalism and the credibility of media outlets using AI tools.

People affected

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

Potential causes

Management

• Pressure to Integrate AI: Corporate management pushed staff to find uses for AI tools.
• Workplace Productivity Pressure: Expectation to keep generating content even while sick.

Technology

• AI Hallucination: ChatGPT generated fabricated quotes instead of verbatim text.
• Experimental Tool Failure: The custom API-based AI tool failed to work as intended.

Data Inputs

• Unverified AI Summaries: Paraphrased AI outputs were used directly without checking original text.

Human Factors

• Working While Severely Ill: Reporter worked from bed with a high fever and very little sleep.
• Overreliance on AI Output: Reporter trusted the AI-generated text without cross-checking.
• Rushing to Meet Deadlines: Pressure to finish quickly led to a critical lack of verification.

Process and Methods

• Inadequate Fact-Checking: Editors failed to verify the quotes against the original source.
• Inconsistent AI Guidelines: Lack of clear, reader-facing guidelines on AI tool usage.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide clear, consistent, and detailed accounts of the incident from multiple sources, including statements from the reporter, the editor-in-chief, and the affected source.

A reporter at Ars Technica used ChatGPT to paraphrase source material, resulting in the publication of fabricated quotes. The incident highlights the risks of generative AI hallucinations and overreliance in professional journalism, but carries negligible national security implications.

• Overall national security impact: Negligible
• Scope: Single nation
• Primary target: United States
• Alleged perpetrator: Individual

Threat characteristics

• Imminence: Long-term. This represents an ongoing, general risk of AI hallucinations in professional environments rather than an active crisis.
• Autonomy: Human-controlled. The AI acted as a writing assistant to a human reporter who retained final control and responsibility for publishing.
• Novelty: Established threat. LLM hallucinations and their accidental inclusion in published texts are well-documented, established risks.

Impact by dimension

• Physical security: Negligible. No physical systems, critical infrastructure, or human safety implications were involved in this localized journalistic error.
• Information security: Negligible. The incident was an isolated journalistic error involving a tech article rather than a coordinated information warfare or intelligence operation.
• Sovereignty: Negligible. No impact on state authority, government functions, sovereign decisions, or electoral systems was reported.
• Economic security: Negligible. The economic impact was limited to individual employment termination and localized media reputational damage, with no strategic economic threat.
• Societal stability: Negligible. No threats to civil liberties, mass human rights violations, or large-scale societal instability occurred.