nabu.science
MethodologyAboutContact

 

Limitation: This content is identified as Perspective / Viewpoint. Evaluation of this content might be partial, and should be treated as indicative.

  • Snapshot
  • Key Claims
  • Strengths & Limitations
  • Commentary
  • From the Authors
  • Author Note / Field Feedback
  • Dimension breakdown
  • Impact Profile
Free · About 5 minutes
Report #NBU-C30640AC·Framework v5.10·Pipeline v0.4·Evaluated 21 Oct 2026·OECD: 6.3 Philosophy, ethics and religion·Subfield: Safety Research·Basis: Full text·Type: Perspective or viewpoint

The Ethics of Prompt Engineering

Germani et al.Germani, Spitale·2026·arXiv (Cornell University)·10.48550/arxiv.2610.23082 ↗

Primary claim

prompting should therefore be understood as an epistemic and normative act

Supported · sustained conceptual reasoning links evidentiary choices and normative framing to user responsibility
In brief

An organized ethical framework connects prompting with epistemic responsibility, but its claim that ethical prompts improve outputs remains untested.

ReliabilityCaveats4 of 4 checks completeRead the entry ↓
Abstract

Prompt engineering is commonly understood as a technical practice aimed at eliciting useful, accurate, and appropriately structured outputs from large language models (LLMs). Yet prompts do more than optimize model performance: they shape what information is elicited, which assumptions are introduced, and which perspectives are privileged. In this paper, we argue that prompting should therefore be understood as an epistemic and normative act. We distinguish between two conceptions of prompting: knowledge generation, in which the model is implicitly treated as an epistemic authority, and knowledge injection, in which users provide relevant knowledge, evidence, and context that the model can organize, articulate, and extend. We contend that the latter provides a more appropriate model for responsible LLM use, since prompting cannot substitute for subject-matter expertise or the critical evaluation of model outputs. At the same time, prompts inevitably inject values, intentions, and assumptions into the generation process. Building on these premises, we propose five principles for responsible prompting: epistemic humility, explicit engagement with uncertainty and epistemic underdetermination, source sensitivity, avoidance of deceptive framing, and recognition of model limitations and appropriate task allocation. We further argue that responsibility for AI-generated outputs is distributed across users, developers, institutions, and societies, and that some applications cannot be rendered ethical through better prompting alone.

Key Claims

Primary finding · Supported

prompting should therefore be understood as an epistemic and normative act

sustained conceptual reasoning links evidentiary choices and normative framing to user responsibility

Abstract
Quote verified
Secondary · Supported

responsibility for AI-generated outputs is distributed across users, developers, institutions, and societies

actor-specific reasoning connects distinct forms of influence with distinct responsibilities

Abstract; The limits of responsible prompting
Quote verified
Secondary · Supported

prompts inevitably inject values, intentions, and assumptions into the generation process

framing examples illustrate how task definitions and desired conclusions embed normative commitments

Abstract
Quote verified
Secondary · Tentative

The ethical improvement is simultaneously an epistemic improvement.

a constructed prompt comparison does not establish improvement in actual model outputs

Ethics as effectiveness
Quote verified

Derived from the full evaluation — not a separate score.

Strengths · Limitations

Strengths

#contribution

Connects epistemic grounding with normative choices

The knowledge-injection and value-injection sections explain how prompts shape both the evidentiary basis and framing of outputs, linking familiar prompting practices into a common ethical account.

↳ Knowledge injection versus knowledge generation; Prompting as value injection

#methodological rigour

Allocates responsibility beyond the individual user

The limits section distinguishes the influence of developers, institutions, users, and societies. It explicitly rejects the idea that better prompting can substitute for system safeguards and institutional accountability.

↳ The limits of responsible prompting: Responsibility is distributed

#reporting

Organizes principles into a traceable argument

The progression from knowledge and value injection to five principles makes the framework easy to follow. The final distinction between instrumental and epistemic effectiveness clarifies what kind of successful prompting the argument concerns.

↳ Core principles of responsible prompting; Ethics as effectiveness

Limitations

#methodological rigour

Knowledge injection lacks a stable boundary

The contrast begins with an unattributed naive model and later includes external retrieval within knowledge injection. This expansion leaves the conceptual distinction less precise than its initial presentation suggests.

↳ Knowledge injection versus knowledge generation

#methodological rigour

Effectiveness claim exceeds its illustrative support

The claim that ethical improvement is also epistemic improvement rests on one constructed prompt comparison. No generated outputs are evaluated, and the definition of epistemic effectiveness already includes ethical features such as avoiding misleading framing.

↳ Ethics as effectiveness

#positioning

Closest frameworks receive limited comparative treatment

The responsible-prompting framework in reference 8 is cited without a substantive comparison of principles or scope. This limits assessment of what the proposed synthesis adds to existing approaches.

↳ An introduction to prompting, reference 8; Core principles of responsible prompting

Does this match your read? Add your assessment ↓

Nabu Commentary

The knowledge- and value-injection sections provide an organized account of why prompting involves more than technical optimization. The distributed-responsibility discussion and treatment of non-experts qualify the practical message and prevent it from becoming a claim of user control over model reliability. The argument is less developed where the injection/generation boundary expands to retrieval and where ethical prompting is said to improve epistemic effectiveness without evaluating outputs. Limited comparison with existing responsible-prompting frameworks further constrains the demonstrated advance, while the principles remain more suitable for teaching and guidance than deployment-ready implementation.

Evaluate another paper

Free · About 5 minutes · Research preview

Author Note / Field Feedback

Nabu’s assessment, alongside the field’s view.

Are you an author of this paper?

Dimension breakdown

Evidence Score

Out of 5
012345
3.2±0.2

Contribution
3.2

The paper integrates knowledge injection, value injection, and distributed responsibility into a discernible conceptual synthesis. Its advance remains incompletely differentiated from the responsible-prompting framework cited as reference 8.

Knowledge injection versus knowledge generation; Prompting as value injection; An introduction to prompting, reference 8
Methodological Rigour
3.1

The argument progresses through examples, principles, and explicit limits, including safeguards for non-experts. The injection/generation boundary is blurred, and the effectiveness assertion relies on an illustrative prompt pair rather than evaluated outputs.

Knowledge injection versus knowledge generation; The limits of responsible prompting; Ethics as effectiveness
Reporting
3.6

The sectional progression and distinction between instrumental and epistemic effectiveness make the main argument traceable. Citation formatting and stronger abstract formulations introduce local imprecision without obscuring the central message.

An introduction to prompting; Abstract; Ethics as effectiveness
Positioning
3.1

Relevant technical and philosophical sources support the discussion, and alternate possibilities concerning personas and non-experts are acknowledged. Comparative engagement with existing responsible-prompting and responsibility frameworks remains limited.

An introduction to prompting; Knowledge injection versus knowledge generation; References, item 8; The limits of responsible prompting

Reliability

Caveats4 of 4 checks


Results checkClean

The reported numbers and statistics are consistent with the stated results.

Reporting and transparencyNoted

References 13, 16, and 30 are listed without corresponding numbered citations in the body. This reporting issue does not change the conceptual conclusions.

Reference IntegrityClean

31 references in manuscript 28 of 29 checkable references found in an index 2 have no canonical index record — counted, but not index-checkable 4 references confirmed by manual review

References not verified automatically are resolved manually within 24 hours.

Post-Publication RecordClean

No retraction or expression of concern found in Crossref (includes Retraction Watch).

Last checked: 21 Oct 2026

Sources: Crossref (incl. Retraction Watch) ✓

Impact Profile

Impact areaPolicy and practice · SDG 16
Type of advanceEmpirical
Route to useNamed
AdvanceAdvances
CitationsToo early
Policy · patentNot yet traced
ReachedNot reachedNot assessable or unverified
ReadinessFoundational
Foundational
Controlled
Piloted
Validated
Implementable
FoundationalImplementable

The primary claim and the five principles are supported only by conceptual argument and illustrative example prompts, with no empirical test or demonstration of the framework working.

From the paper
“Institutions must establish policies, provide training, and create accountability structures that ensure AI is used appropriately within their contexts.”

The paper names institutions such as hospitals, universities and news organisations, and users like physicians, as actors who would apply its principles through policies, training and accountability structures.

Pathway

Institutions deploying LLMs, such as hospitals, universities and newsrooms, and their users could adopt the five principles in AI-use policies, training and prompting practice.

Barrier

The framework rests on conceptual argument and illustrative examples alone, with no empirical test, operational measures or evidence that following the principles improves outputs.

Evaluated 13 Nov 2026 · GBD 2023 · glm_v1_3_3 · impact_profile_v1_3_3

Report an issue →

AI-generated, human-governed. Something look off? Contact us to request a review.

Keep reading

Related evaluations

Other papers Nabu has evaluated blind, starting with the same field.

  1. When AI Reviews Train AI Reviewers: Scientific-Judgment Collapse and Mitigation

    Ho, Liu, Tran, et al. · 2026 · Computer and information sciences

    Primary claim

    After one recursive step, increasing exposure from 0% to 100% makes recommendation ratings more concentrated, independently generated reviews of the same paper about 11% more similar, and reviews across different papers about 5% less diverse.

    Supported
    Evidence Score
    3.6±0.2
    Reliability
    No concerns
  2. Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender

    Steven D Shaw, Gideon Nave · 2026 · Psychology

    Primary claim

    Relative to baseline (no System 3 access), accuracy significantly rose when AI was accurate and fell when it erred (+25/-15 percentage points; Study 1), the behavioral signature of cognitive surrender

    Supported
    Evidence Score
    3.6±0.2
    Reliability
    Caveats
  3. Neighborly social pressure and collective action: Evidence from a field experiment in Tunisia

    Jöst · 2024 · Political science

    Primary claim

    I do not find evidence to suggest that neighborly social pressure encourages participation in neighborhood cleanups, with low participation rates both for the neighbor and outsider contact conditions.

    Tentative
    Evidence Score
    3.2±0.2
    Reliability
    Caveats

How the rubric works →