FAccT-Checked: A Narrative Review of Authority Reconfigurations and Retention in AI-Mediated Journalism

1. Problem Statement

The integration of Generative AI (specifically Large Language Models or LLMs) into journalistic workflows has intensified concerns regarding Fairness, Accountability, and Transparency (FAccT). However, current discourse often treats these concerns as purely technical properties of AI systems (e.g., model bias or explainability).

The authors argue that FAccT issues in journalism cannot be understood in isolation from editorial authority. As AI tools are adopted, the legitimate power to define, describe, and explain reality (editorial authority) is being reconfigured. Without addressing these structural shifts in power, FAccT principles become difficult to maintain, accountability becomes unassignable, and transparency becomes performative. The core problem is that authority is migrating away from journalists to both internal AI systems and external technology vendors, creating "responsibility gaps" and obscuring the locus of decision-making.

2. Methodology

The paper employs a Critical Narrative Review approach, synthesizing literature from three distinct disciplinary fields:

• Journalism Studies: Focusing on gatekeeping, professional autonomy, and epistemic legitimacy.
• Human-Computer Interaction (HCI): Focusing on human-AI interaction, automation bias, and participatory design.
• FAccT Scholarship: Focusing on algorithmic accountability, fairness, and transparency.

Process:

• Search Strategy: A structured search across major databases (Scopus, ACM, IEEE, Web of Science, JSTOR, Google Scholar) and grey literature (1955–2025).
• Screening: An iterative four-phase process:
  1. Identifying FAccT concerns in AI-mediated journalism.
  2. Discovering "authority" as the underlying structural condition.
  3. Refining the corpus around editorial authority and judgment.
  4. Expanding to include participatory methods and platform dependencies.
• Corpus: The final review analyzed 209 records, including empirical studies, theoretical frameworks, and practitioner accounts.
• Analytical Framework: The authors conceptualize editorial authority as the conjunction of three dimensions:
  1. Decision Rights: Who chooses what gets published?
  2. Epistemic Warrant: Whose knowledge claims are recognized as legitimate?
  3. Responsibility: Who can be held accountable for outcomes?

3. Key Contributions

The paper makes four primary contributions:

1. A Holistic Framework for Editorial Authority: It synthesizes insights to define authority not just as a role, but as a structural precondition for FAccT guarantees.
2. Documentation of Internal Authority Migration: It details how editorial judgment is progressively deferred to LLMs through cognitive and social mechanisms within the newsroom.
3. Mapping of External Authority Migration: It traces how decision-making power shifts from newsrooms to platforms, vendors, and infrastructure providers.
4. Critical Evaluation of Participatory Methods: It assesses the potential of participatory design to reclaim authority while highlighting the risks of "pseudo-participation."

4. Key Results and Findings

A. Internal Migration of Authority (Newsroom to LLM)

Authority is shifting from journalists to AI systems through subtle, non-policy-driven mechanisms:

• Cognitive Mechanisms: Automation bias leads journalists to defer to AI outputs, assuming they are more objective. Anthropomorphism and the CASA (Computer as Social Actors) paradigm cause users to treat LLMs as colleagues rather than tools.
• Social Legitimacy: LLMs exhibit institutional voice mimicry, reproducing journalistic norms (e.g., passive voice, "according to sources") which creates a "strong" perceived alignment with professional values, despite "weak" technical alignment (RLHF optimization for plausibility rather than truth).
• Responsibility Gaps: The opacity of LLM reasoning creates pseudo-transparency. Explanations (e.g., "this serves the public interest") act as "moral buffers," allowing journalists to deflect responsibility. The result is a shift where journalists become validators rather than authors, retaining formal accountability but losing control over the epistemic process.

B. External Migration of Authority (Newsroom to Platforms/Vendors)

Decision-making power is shifting upward to technology providers:

• Infrastructural Capture: Newsrooms depend on external platforms for distribution, monetization, and now production tools (LLMs). This creates asymmetries where a small group of tech companies controls the datasets and models.
• Differential Vulnerability: Small and local newsrooms are disproportionately affected, lacking the resources to negotiate terms or build independent AI capabilities, leading to a concentration of power among large conglomerates.
• Feedback Loops: Using platform tools generates data that further trains the platforms' models, reinforcing their dominance and embedding external values into editorial processes.

C. Participatory Methods as Retention Mechanisms

Participatory design offers a potential counter-force but is limited by structural constraints:

• Degrees of Participation: Moving from consultation (tokenistic) to collaboration and ownership is necessary to redistribute authority.
• The "Participatory Ceiling": Journalists often face limits in influencing "one-size-fits-all" commercial models.
• Conditions for Success: Meaningful retention of authority requires early intervention (upstream in design), collective rather than individual action, and formal institutional support. Without these, participation can become extractive or legitimize existing power imbalances.

5. Significance and Implications

• Reframing FAccT: The paper argues that FAccT interventions cannot succeed through technical optimization alone (e.g., better explainability). Instead, they must address the institutional power relations and epistemic legitimacy that define who holds authority.
• Policy and Practice: It calls for a shift in how AI is integrated into newsrooms. Rather than treating AI as a neutral tool, organizations must explicitly map and govern the redistribution of decision rights, epistemic warrant, and responsibility.
• Future Research: The authors advocate for empirical studies that trace authority reconfigurations across different organizational contexts, moving beyond technical evaluations to examine the socio-political dynamics of AI adoption in journalism.
• Supplementary Assets: The paper provides a Practitioner Concerns Atlas (Table A) and a comprehensive AI Journalism Atlas (Table B) mapping specific AI functionalities across the journalistic workflow, serving as a foundational resource for future research and policy.

In conclusion, the paper posits that the crisis of FAccT in AI-mediated journalism is fundamentally a crisis of authority. Unless the migration of editorial power to algorithms and platforms is actively managed and contested, the core democratic functions of journalism—fairness, accountability, and transparency—will remain compromised.