It Is Not My Code Anymore
cs.SE · cs.AI
Quick verdict
Neat position note: who produced defective LLM code ≠ who is responsible; acceptance must be demand-driven. Clear but purely discursive, no empirical support.
Summary
This research note examines authorship and responsibility in AI-assisted programming through a hypothetical scenario: an LLM-generated division-by-zero expression, approved by an inattentive human reviewer, causes a student's enrollment to fail. The author argues that identifying who produced a defective expression does not by itself determine the duties of reviewers, release managers, or operators, and that psychological ownership, authorship claims, and post-failure responsibility are separable judgments — drawing on studies of AI contribution and perceived ownership (Seo et al., 30 participants; Draxler et al.'s AI Ghostwriter Effect), Naur's theory building, and responsibility-gap philosophy (Elish; Santoni de Sio and Mecacci). Framing generative models as 'machine-in-the-loop' within an existing machine-based process, the note borrows from quality engineering history (Shewhart's 1924 control charts, FDA process validation) to argue for a central quality-engineering role defining acceptance evidence. It distinguishes inside-out from outside-in evaluation, where demand-derived criteria (e.g., a valid enrollment record, no duplicate charges; Table 1) judge both individual deliveries and repeated production runs. It closes by asking whether tasks need separately generated programs at all, contrasting Hassan et al.'s SE 3.0 and Welsh's 'End of Programming.' Limitations are candid: the example is hypothetical, no new empirical results are reported, and the literature reading is selective. The value is conceptual clarity; the cost is that all normative recommendations rest on argument rather than evidence.
Review complete
Assessment at a glance
3
risk flags identified
● high 0
● medium 2
● low 1
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 18 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Identifying the producer of a defective expression does not by itself determine the duties of reviewers, release decision-makers, or service operators.
- Feeling ownership, claiming authorship, and accepting responsibility are separable judgments, so collective ownership does not resolve post-failure responsibility gaps.
- Acceptance criteria must be justified by the required service outcome (outside-in), with component checks contributing evidence rather than serving as the oracle.
- Quality engineering should hold a central role with explicit duties and decision authority over both generated artifacts and the production process.
- If an existing trained system performs a task without generating a separate program, the object of authorship disappears while the service obligation remains.
Strengths
Explicit honesty about scope
The abstract, footnote, and Limitations section all state the enrollment failure is hypothetical, no new empirical results are reported, and the literature reading is selective — a rare level of overclaim-avoidance in the vibe-coding discourse.
Genuine conceptual decomposition
The Authorship section cleanly separates producer, reviewer, maintainer, and owner via the 'my ride' analogy, grounded in Seo et al.'s 30-participant finding (attribution shifts toward AI while accountability acceptance stays stable) and Draxler et al.'s AI Ghostwriter Effect.
Actionable evaluation framing
Productive historical anchoring
Forward-looking closing question
Weaknesses
Zero empirical content
Normative assertions without operationalization
Self-citation dependency
Selective literature with consequential gaps
Single, morally transparent running example
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This is a self-declared research note — the abstract and the Limitations section both state that it reports no new empirical results — spanning authorship, responsibility, and evaluation in AI-assisted software production. Its running example is deliberately hypothetical: an LLM introduces `1/0` while implementing a feature, an inattentive developer approves the diff, and a student's enrollment fails. The central claims are: (1) producer identity is orthogonal to duty allocation — 'Reviewing it did not change who produced it,' and knowing the producer 'does not tell us who should have caught the defect or who must restore the service'; (2) ownership feeling, authorship claim, and responsibility acceptance are separable judgments; (3) acceptance criteria should be justified by the required service outcome, not by the producing process; (4) quality engineering should hold a central organizational role with explicit duties and decision authority; (5) substituting a trained system for a generated program removes the object of authorship but not the service obligation. The claims are conceptual, and the paper is scrupulous about not presenting them as empirical findings.
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 3 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the paper 'It Is Not My Code Anymore' about?
It is a 2026 arXiv research note by Augusto Camargo arguing that when an AI model writes code, identifying who produced a defective line does not by itself determine who must review it, who approves release, who fixes failures, or who answers for harm. It proposes evaluating AI-assisted software against the required service outcome (e.g., 'the student must be able to enroll') rather than only against code-level checks.
Does the paper report experiments or data?
No. The author explicitly states the paper reports no new empirical results. It uses one hypothetical scenario — an LLM-generated division-by-zero crash, approved by an inattentive reviewer, that breaks student enrollment — plus a selective reading of published studies on code ownership and responsibility.
What is the 'moral crumple zone' mentioned in the paper?
A concept from Madeleine Clare Elish (2019) describing how a nearby human operator with limited actual control can absorb blame after an automated system fails. The paper cites it to warn that blaming the developer who approved an AI-written diff may misassign responsibility for defects the model introduced.
What does the paper conclude about who is responsible when AI code fails?
That assigning responsibility requires three things: a relevant contribution, an applicable duty, and evidence connecting them. Producer identity, psychological ownership, and post-failure duties are separate questions with potentially different answers — the model produced the expression, the production process accepted it, and the operator owed the student a working service.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figure 1 is an image; only its caption is present in the extracted text, so the satirical figure content itself cannot be assessed.
- Figure 2's diagram labels are extracted in raw reading order, so the arrow structure and exact label-to-flow mapping are approximate.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.