Self-Admitted GenAI Usage in Open-Source Software

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Recent research has shown that using AI-generated PR descriptions reduces review time and increases PR merge rates.

“Recent research has shown that using AI-generated PR descriptions reduces review time and increases PR merge rates [7].”

• cites [7] sourceSupported

May combine two claims.

Developers perceived productivity improvements when using GitHub Copilot, particularly for repetitive and routine coding activities, with the perceived magnitude of improvement varying considerably based on task complexity and context.

“Based on surveys and usage telemetry, they found that developers perceived productivity improvements when using Copilot, particularly for repetitive and routine coding activities, with the perceived magnitude of improvement varying considerably based on task complexity and context.”

• cites [4] sourcePartly supported

  The Productivity survey and Conclusions report perceived productivity gains, including faster repetitive work and help with routine tasks. The paper also reports differences by developer experience, but the passages do not establish that the magnitude of perceived improvement varies considerably with task complexity and context. (LLM, conf=0.83)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Asare et al. found that while the GenAI tool introduced security vulnerabilities, the rates were not higher than those introduced by human developers.

“Asare et al. [43] compared vulnerability rates between human-written and Copilot-generated code and found that, while the GenAI tool introduced security vulnerabilities, the rates were not higher than those introduced by human developers.”

• cites [43] sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Pearce et al. found systematic security weaknesses in AI-generated code across multiple programming languages and contexts.

“Pearce et al. [53] conducted a security assessment of code contributions generated by GitHub Copilot across multiple programming languages and contexts.They found systematic security weaknesses in AI-generated code”

• cites [53] sourceSupported

Classified as probably not a substantial claim.

Previous research links code clones to increased technical debt.

“and code clones to increased technical debt [55,56].”

• cites [55] Aversano, 2019, Investigating on the impact of software clones on technical debtNot checked yet

  Not checked in the free run.

• cites [56] sourcePartly supported

  The Interpretation of Results and Practical Considerations section reports that reused code has about three times the technical-debt principal of native code. However, the paper studies reused classes, not code clones specifically, and its Abstract reports lower technical-debt interest for reused code. It therefore supports a narrower link between code reuse and one measure of technical debt, not the claim about clones generally. (LLM, conf=0.76)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

Pearce et al. argued that security issues introduced by GenAI tools stem from the models' training on publicly available code repositories, which inherently contain security flaws.

“arguing that security issues introduced by GenAI tools stem from the models' training on publicly available code repositories, which inherently contain security flaws.”

• cites [53] sourcePartly supported

  The Introduction says Copilot was trained on open-source code that includes insecure coding patterns, creating the potential to generate code with those patterns. The claim extends that explanation from Copilot to GenAI tools generally and presents training data as the established cause of security issues; the paper frames it as a potential mechanism, not a demonstrated cause. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Generated messages are not limited to this one project; similar patterns appear in popular projects such as pytorch/pytorch and hasura/graphql-engine.

“These generated messages are not limited to this one project-similar patterns appear in popular projects such as pytorch/pytorch and hasura/graphql-engine.”

May combine two claims.

One separate case was identified in which contributors were encouraged to use ChatGPT to produce pull request descriptions (P11); this case was not part of the Copilot for Pull Requests group.

“In addition, we identified one separate case in which contributors were encouraged to use ChatGPT to produce pull request descriptions (P11); this case was not part of the Copilot for Pull Requests group.”

May combine two claims.

Running the retrieval process on all repositories yielded 3,004 mentions of GenAI tools: 1,572 in commit messages, 397 in source code comments, and 1,035 in documentation files.

“Running the above retrieval process on all repositories yielded 3,004 mentions of GenAI tools: 1,572 in commit messages, 397 in source code comments, and 1,035 in documentation files.”

May combine two claims.

In other cases, changes were reverted (1), AI-generated comments were deleted (2), or the generated code was commented out (1).

“In other cases, changes were reverted (1), AI-generated comments were deleted (2), or the generated code was commented out (1).”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The linked contribution guidelines (E6) do not discuss GenAI usage.

“Interestingly, the linked contribution guidelines (E6) do not discuss GenAI usage.”

May combine two claims. The quote only partly states this claim.

In one commit message (E1), the developer acknowledged that the code was written 'a bit hasty on previous release' due to 'trust in GitHub Copilot'; the developer task was bug fixing, while the initial GenAI task was code generation.

“in one commit message that we analyzed (E1) the developer acknowledged that the code was written”

May combine two claims.

Excluding PR-related activities, generation tasks were most common, with code generation being particularly prominent (105 instances).

“Unsurprisingly, excluding PR-related activities, generation tasks were most common, with code generation being particularly prominent (105 instances).”

Classified as probably not a substantial claim.

The study conducted a developer survey to uncover the ethical, legal, and practical concerns behind GenAI policies.

“conduct a developer survey to uncover the ethical, legal, and practical concerns behind them”

Probably not a substantial claim: may not pass S3 (load-bearing).

This approach was very common in one particular project, which contributed 1,000 such mentions to the sample.

“this approach was very common in one particular project, which contributed 1,000 such mentions to our sample.”

May combine two claims.

Translation followed with 50 instances, while optimization and maintenance tasks accounted for 34 and 26 instances, respectively.

“Translation followed with 50 instances, while optimization and maintenance tasks accounted for 34 and 26 instances, respectively.”

Probably not a substantial claim: may not pass S3 (load-bearing).

In false positive matches, mentions of GenAI tools were often surrounded by commas or underscores, e.g., when they were part of URLs for API calls.

“we noticed that in false positive matches, the mentions of GenAI tools were often surrounded by commas or underscores, e.g., when they were part of URLs for API calls.”

May combine two claims.

Of the 1,008 Copilot for Pull Requests mentions, 1,000 occurred in a single repository (pancakeswap/pancake-frontend) where commit messages directly reused pull request descriptions generated by the tool.

“Of these, 1,000 instances occurred in a single repository (pancakeswap/pancake-frontend) where commit messages directly reused pull request descriptions generated by the tool”

Probably not a substantial claim: may not pass S3 (load-bearing).

Developers cited concerns such as copyright issues, license violations, and ethical considerations as key reasons for establishing guidelines.

“They cited concerns such as copyright issues, license violations, and ethical considerations as key reasons for establishing guidelines.”

May combine two claims.

Of the 151 repositories with self-admitted GenAI usage, 119 had a significant difference in file-based churn, and 120 had a significant difference in line-based churn (p < 0.05).

“Of the 151 repositories with self-admitted GenAI usage, 119 had a significant difference in file-based churn, and 120 had a significant difference in line-based churn (p < 0.05).”

May combine two claims.

Survey participants mentioned aspects including the unclear copyright situation of training data, unclear implications for generated content, data privacy risks when sharing inputs with GenAI systems, and concerns regarding code quality and maintainability.

“Mentioned aspects include the unclear copyright situation of the training data, the unclear implications for generated content, data privacy risks when sharing inputs with GenAI systems, and concerns regarding code quality and maintainability.”

May combine two claims.

Acknowledgment of GenAI usage was a common purpose, sometimes combined with warnings about possible negative implications.

“Acknowledgment of GenAI usage was a common purpose, sometimes combined with warnings about possible negative implications.”

May combine two claims.

The average line-based code churn over all commits decreased from 0.68 to 0.50 pre- to post-GenAI adoption, with a significant difference (p < 0.05) and a negligible effect (|δ| = 0.09).

“the average line-based churn decreased from 0.68 to 0.50 with a significant difference (p < 0.05) and a negligible effect (|δ| = 0.09).”

Probably not a substantial claim: may not pass S3 (load-bearing).

Some developers indicated that they rely on automated tools to evaluate generated content.

“Additionally, some developers indicated that they rely on automated tools, e.g., code quality checks or automated testing, to evaluate generated content.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.

D6 added that disclosing the percentage of GenAI involvement in contributions could reduce the likelihood of generated noise PRs and improve code review efficiency.

“D6 added that disclosing the percentage of GenAI involvement in contributions could reduce the likelihood of generated”

Probably not a substantial claim: may not pass S3 (load-bearing).

D5 pointed to ethical and legal ambiguities related to AI-generated code, describing them as maintainer's main concerns.

“D5, pointed to”

Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based). May combine two claims.

The associated recommendation emphasizes that GenAI tools often rely on reinforcement learning, which could expose sensitive data to unintended parties.

“The associated recommendation emphasizes that GenAI tools often rely on reinforcement learning, which could expose sensitive data to unintended parties.”

Probably not a substantial claim: may not pass S3 (load-bearing).

D5,6,8 emphasize the importance of documenting prompts and generated content to ensure accountability in software projects.

“Documenting Prompts and Generated Content: D5,6,8 emphasize the importance of documenting prompts and generated content to ensure accountability in software projects.”

Probably not a substantial claim: may not pass S3 (load-bearing).

D4 opposed such restrictions, viewing them as unnecessary limitations that could stifle productivity and innovation.

“D4 opposed such restrictions, viewing them as unnecessary limitations that could stifle productivity and innovation.”

May combine two claims.

A decreasing churn was more common: 49 repositories had a decreasing file-based churn with a medium-to-large effect size, and 49 repositories had a decreasing line-based churn with a medium-to-large effect size.

“A decreasing churn was more common: 49 repositories had a decreasing file-based churn with a medium-to-large effect size, and 49 repositories had a decreasing line-based churn with a mediumto-large effect size.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

D4 and D6's feedback on P4 emphasizes concerns about the high false positive rates of AI identification tools and warns against deferring critical decisions to automation.

“D4 and D6's feedback on P4, which regulates the use of AI in community discussions, emphasizes concerns about the high false positive rates of AI identification tools and warns against deferring critical decisions to automation.”

May combine two claims.

Eleven repositories had an increasing file-based churn with a medium-to-large effect size, and 15 had an increasing line-based churn with a medium-to-large effect size.

“Eleven repositories had an increasing file-based churn with a medium-to-large effect size, and 15 had an increasing linebased churn with a medium-to-large effect size.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The mentions were grouped into three groups: prohibitive, restrictive, and supportive usage.

“We grouped them into three groups: (1) prohibitive, (2) restrictive, and (3) supportive usage.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The project owner of jqwik-team/jqwik described their decision to disallow the use of GenAI tools as an ethical decision due to all its collateral damages.

“For example, the project owner of jqwik-team/jqwik (D3) described their decision to disallow the use of GenAI tools as an”

May combine two claims.

For the 1,292 GenAI mentions analyzed, developers mainly used GenAI tools for code generation, natural language translation, and code refactoring.

“For the 1,292 GenAI mentions we analyzed, developers mainly used GenAI tools for code generation, natural language translation, and code refactoring.”

Probably not a substantial claim: may not pass S3 (load-bearing).

D5 specifically highlighted that AI models are likely trained on large datasets that include open-source codebases with various licensing terms.

“They specifically highlighted that”

May combine two claims.

The project spring-projects/spring-cli promoted the use of GenAI to generate README.md files and has developed GenAI tooling to support automated documentation rewriting.

“The project spring-projects/spring-cli (P13) promoted the use of GenAI to generate README.mdfiles and has even developed GenAI tooling to support automated documentation rewriting.”

May combine two claims.

Our analysis revealed that developers primarily use GenAI tools for code generation, natural language translation, and code refactoring.

“Our analysis revealed that developers primarily use GenAI tools for code generation, natural language translation, and code refactoring.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

D5 argued that while LLMs are suitable for repetitive tasks and generic translations, they lack the creativity needed for meaningful contributions.

“D5 argued that while LLMs are suitable for repetitive tasks and generic translations, they lack the creativity needed for meaningful contributions.”

Probably not a substantial claim: may not pass S3 (load-bearing).

When asked about specific aspects of software projects that should be regulated, developers expressed concerns primarily about unlicensed training datasets and potential licensing conflicts associated with AI-generated code.

“When asked about specific aspects of software projects that should be regulated, developers expressed concerns primarily about unlicensed training datasets and potential licensing conflicts associated with AI-generated code.”

Probably not a substantial claim: may not pass S3 (load-bearing).

D2 explicitly stated that no additional actions are necessary.

“Interestingly, D2 explicitly stated that no additional actions are necessary, explaining that”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

D5 expressed a negative view of regulating GenAI tool usage, arguing that it could hinder productivity.

“Interestingly, D5 expressed a negative view of regulating GenAI tool usage, arguing that it could hinder productivity.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The regulation in the accompanying contributor agreement (P1) explicitly prohibits contributions created using GenAI tools, citing the unresolved legal implications of training AI models on public code repositories.

“The regulation in the accompanying contributor agreement (P1) explicitly prohibits contributions created using GenAI tools, citing the unresolved legal implications of training AI models on public code repositories.”

Probably not a substantial claim: may not pass S3 (load-bearing).

D7 supported the regulation, noting its alignment with their own concerns about the ethical and legal implications of using GenAI tools.

“D7 supported the regulation, noting its alignment with their own concerns about the ethical and legal implications of using GenAI tools.”

May combine two claims.

Quality assurance emerged as another key purpose, with developers often requesting peer review of AI-generated content.

“Quality assurance emerged as another key purpose, with developers often requesting peer review of AI-generated content.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Participants argued for transparent disclosure of GenAI usage and for documenting the generation context.

“In relation to that, participants argued for transparent disclosure of GenAI usage and also for documenting the generation context.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The project owner of owasp/wrongsecrets focuses on the ethical risks of using GenAI when describing the rationale behind guideline P8.

“Similarly, the project owner of owasp/wrongsecrets (D2) focuses on the ethical risks of using GenAI when describing the rationale behind guideline P8.”

May combine two claims.

Despite some developers considering prompt documentation unnecessary, the majority agreed that it is valuable to understand the extent of GenAI's contributions to a project.

“Despite some developers considering prompt documentation unnecessary, the majority agreed that it is valuable to understand the extent of GenAI's contributions to a project.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Three developers highlighted these practices as crucial steps to ensure the quality and compliance of GenAI-based contributions.

“Three developers highlighted these practices as crucial steps to ensure the quality and compliance of GenAI-based contributions.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The maintainers proposed to incorporate indicators for identifying possible GenAI usage and suggested tool support.

“In addition, the maintainers proposed to incorporate indicators for identifying possible GenAI usage and suggested tool support, for example, a ChatGPT detector published on HuggingFace [21].”

• cites [21] Guo, 2023, How close is ChatGPT to human experts? Comparison corpus, evaluation, and detection sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

One mention referred to translation between programming languages.

“One mention referred to translation between programming languages.”

May combine two claims.

We found 13 policies and guidelines on GenAI usage in open-source software projects, including strict policies prohibiting GenAI usage, policies requiring attribution, but also guidelines encouraging contributors to use GenAI.

“We found 13 policies and guidelines on GenAI usage in open-source software projects, including strict policies prohibiting GenAI usage, policies requiring attribution, but also guidelines encouraging contributors to use GenAI, for example, for translating natural language text.”

May combine two claims.

The average file-based code churn over all commits decreased from 0.17 to 0.06 pre- to post-GenAI adoption, with a significant difference (p < 0.05) and a medium effect (|δ| = 0.42).

“The average file-based code churn decreased from 0.17 to 0.06 with a significant difference (p < 0.05) and a medium effect (|δ| = 0.42)”

May combine two claims. The quote only partly states this claim.

For the 149 repositories included, we identified four patterns of code churn trends after GenAI adoption: upward trend with positive slope change, upward trend with negative slope change, downward trend with positive slope change, and downward trend with negative slope change.

“For the 149 repositories included, we identified four patterns, that we describe in the following:”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

D6 criticized the policy as being overly cautious, suggesting that asking contributors to disclose percentage of generated content is sufficient.

“D6 criticized the policy as being overly cautious, suggesting that asking contributors to”

Probably not a substantial claim: may not pass S3 (load-bearing).

Opinions diverge considerably for P3, which prohibits AI-generated code.

“Opinions diverge considerably for P3, which prohibits AIgenerated code.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Five developers highlighted the necessity of regulating the usage of GenAI tools in software projects.

“Five developers highlighted the necessity of regulating the usage of GenAI tools in software projects.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Table 7 lists 13 software artifacts from 12 GitHub repositories that presented policies or usage guidelines for GenAI usage.

“As mentioned above, Table 7 lists 13 software artifacts from 12 GitHub repositories that presented policies or usage guidelines for GenAI usage.”

May combine two claims.

The remaining nine cases include eight instances in which developers explicitly documented the use of Copilot for Pull Requests to generate PR descriptions, and one instance suggesting that contributors use ChatGPT to produce PR descriptions (P11).

“The remaining nine cases include eight instances in which developers explicitly documented the use of Copilot for Pull Requests to generate PR descriptions (Section 3.1), and one instance suggesting that contributors use ChatGPT to produce PR descriptions (P11).”

May combine two claims.

The analysis identified 11 different purposes for GenAI mentions in software projects, grouped into four main categories.

“Our analysis identified 11 different purposes for GenAI mentions in software projects, grouped into four main categories (see Table 6).”

May combine two claims.

Maintainers are already actively regulating GenAI usage, but through heterogeneous approaches ranging from bans to selective encouragement.

“RQ2 shows that maintainers are already actively regulating GenAI usage, but through heterogeneous approaches ranging from bans to selective encouragement.”

May combine two claims.

The RQ3 results show no systematic increase in code churn after GenAI adoption, contradicting prominent industry claims.

“The RQ3 results show no systematic increase in code churn after GenAI adoption, contradicting prominent industry claims.”

May combine two claims.

Our study analyzes self-admitted GenAI usage across more than 200,000 OSS repositories and conducts a study of code churn, showing how OSS projects use GenAI tools and how their usage impacts development activity.

“Our study adds to the existing body of knowledge by analyzing self-admitted GenAI usage across more than 200,000 OSS repositories and conducting a study of code churn, showing how OSS projects use GenAI tools and how their usage impacts development activity.”

May combine two claims.

The RQ1 results show that developers most often use GenAI for code generation, natural-language translation, and refactoring-with explicit acknowledgment as the dominant purpose.

“The RQ1 results show that developers most often use GenAI for code generation, natural-language translation, and refactoring-with explicit acknowledgment as the dominant purpose.”

Classified as probably not a substantial claim.

This study introduced self-admitted GenAI usage-explicit references to LLM-based tools such as ChatGPT and GitHub Copilot-as a novel lens for examining how generative AI is used in open-source software development.

“This study introduced self-admitted GenAI usage-explicit references to LLM-based tools such as ChatGPT and GitHub Copilot-as a novel lens for examining how generative AI is used in open-source software development.”

May combine two claims.

While some repositories exhibit increased churn, many show stable or decreasing trends, and effects vary widely across projects.

“While some repositories exhibit increased churn, many show stable or decreasing trends, and effects vary widely across projects.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Three authors collaborated throughout the coding process to ensure rigorous and consistent annotation.

“Three authors of this paper collaborated throughout this process to ensure a rigorous and consistent annotation.”

May combine two claims.

The finding of no general increase in code churn contradicts popular narratives on the impact of GenAI on software development.

“contradicting popular narratives on the impact of GenAI on software development”

Classified as probably not a substantial claim.

A sample size of 341 mentions (of 3,004) was calculated to achieve estimates with a 95% confidence level and a 5% confidence interval.

“We calculated a sample size of 341 mentions (of 3,004) to achieve estimates with a 95% confidence level and a 5% confidence interval.”

May combine two claims.

ChatGPT was the most popular general AI tool and GitHub Copilot the most popular AI developer tool among developers as of the 2023 Stack Overflow Developer Survey.

“In that survey, ChatGPT was identified as the most popular general AI tool and GitHub Copilot as the most popular AI developer tool.”

• cites [16] 2023, Stack overflow developer survey 2023Not checked yet

  Not checked in the free run.

May combine two claims.

Inter-rater reliability for task, content, and purpose coding yielded 'substantial' to 'almost perfect' agreement levels: task (k = 0.81 -0.89), content (k = 0.95 -0.99), and purpose (k = 0.79 -0.92).

“The analysis yielded”

• cites [20] Viera, 2005, Understanding interobserver agreement: The kappa statistic, Family Med.Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

There are reports of 'AI slop' wasting valuable time of maintainers or GenAI-generated contributions leading to more rework.

“there are also reports of”

• cites [10] StenbergNot matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [11]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

This practice represents a form of explicit, self-admitted GenAI usage, embedding a clear marker of AI contribution directly into the software project's official history.

“This practice represents a form of explicit, self-admitted GenAI usage, embedding a clear marker of AI contribution directly into the software project's official history.”

May combine two claims.

Filtering out repositories with a code ratio outside the 97% confidence interval helps eliminate outliers, that is, repositories with very little code, or codebases dominated by code without comments.

“Filtering out repositories beyond the 97% confidence interval helps eliminate outliers, that is, repositories with very little code, or codebases dominated by code without comments.”

Classified as probably not a substantial claim.

Two authors independently labeled a sample of mentions to evaluate the coding instructions.

“To evaluate the coding instructions, two authors independently labeled a sample of mentions, deciding whether they should be included or not.”

Classified as probably not a substantial claim.

The 1,008 Copilot for Pull Requests mentions were set aside during the initial round of coding to avoid skewing the development of the coding schema.

“we set aside these 1,008 mentions during the initial round of coding to avoid skewing the development of the coding schema.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Two authors independently analyzed 284 GenAI mentions to build the code book.

“To build the code book, two authors independently analyzed 284 GenAI mentions.”

May combine two claims.

Although the single use case contributes a large absolute count, it represents only one entry in the taxonomy of GenAI tasks and does not affect the broader findings.

“although this single use case contributes a large absolute count, it represents only one entry in our taxonomy of GenAI tasks (Table 3) and does not affect our broader findings.”

Probably not a substantial claim: may not pass S2 (checkable).

OSS projects, with their collaborative nature and publicly accessible repositories, offer a unique context to study the adoption of GenAI tools.

“These projects, with their collaborative nature and publicly accessible repositories [9], offer a unique context to study the adoption of GenAI tools.”

• cites [9] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

While GitClear used a global cutoff date, we used the first GenAI mention in a repository as a proxy for GenAI adoption, thus following a more fine-grained approach.

“While GitClear used a global cutoff date, we used the first GenAI mention in a repository as a proxy for GenAI adoption, thus following a more fine-grained approach.”

• AssumptionThe authors state this as an assumption: The first GenAI mention in a repository is a valid proxy for the time of GenAI adoption.

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

This aligns with the cautionary tone of the regulation, which warns about overreliance on AI-generated content.

“This aligns with the cautionary tone of the regulation, which warns about overreliance on AI-generated content.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

These policies and usage guidelines reflect a growing awareness of both the opportunities and risks of GenAI tools in open-source software projects and the willingness of the maintainers to guide their usage.

“Overall, these policies and usage guidelines reflect a growing awareness of both the opportunities and risks of GenAI tools in open-source software projects and the willingness of the maintainers to guide their usage.”

Classified as probably not a substantial claim.

Code churn serves as an indicator of the maturity of the code that developers add or modify.

“Therefore, it serves as an indicator of the maturity of the code that developers add or modify.”

Probably not a substantial claim: may not pass S3 (load-bearing).

GitClear analyzed over 150 million lines of code across GitHub repositories from 2020 to 2023 to assess the impact of AI-assisted development on code quality.

“In 2024, GitClear analyzed over 150 million lines of code across GitHub repositories from 2020 to 2023 to assess the impact of AI-assisted development on code quality [11]”

• cites [11]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The feedback from open-source developers regarding GenAI tool guidance reflects a combination of ethical, legal, and practical considerations.

“The feedback we received from open-source developers regarding GenAI tool guidance reflects a combination of ethical, legal, and practical considerations.”

May combine two claims.

While some repositories have an increasing trend in code churn, both the overall trend and the trend in many individual repositories point to a decreasing code churn over time.

“While we observed that some repositories have an increasing trend in code churn, both the overall trend and the trend in many individual repositories point to a decreasing code churn over time.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Files affected by technical debt tend to be more defect-prone.

“files affected by technical debt tend to be more defect-prone [24].”

• cites [24] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

After applying these thresholds, we excluded two repositories with insufficient data.

“After applying these thresholds, we excluded two repositories with insufficient data.”

Classified as probably not a substantial claim.

Code churn is a widely recognized indicator of software maintainability.

“Code churn is a widely recognized indicator of software maintainability [22].”

• cites [22] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The GitClear 2025 report documented an eight-fold increase in duplicated code blocks during 2024.

“The same report documented an eight-fold increase in duplicated code blocks during 2024”

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The GitClear 2025 report reported a 17.1% increase in copy-pasted code.

“a 17.1% increase in copy-pasted code”

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The 2024 GitClear report suggested that code churn has been continuously increasing since the adoption of GenAI tools in software projects.

“The 2024 GitClear report [11] suggested that code churn has been continuously increasing since the adoption of GenAI tools in software projects.”

• cites [11]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim. May combine two claims.

GitClear attributed rising code churn to GenAI adoption and interpreted it as indicative of code that was incomplete or erroneous when initially committed.

“attributing rising code churn to GenAI adoption and interpreting it as indicative of code that was incomplete or erroneous when initially committed.”

• cites [11]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

This documentation is essential for assessing which code is potentially affected by copyright and licensing issues; it might also prove useful for later maintenance activities.

“This documentation is essential for assessing which code is potentially affected by copyright and licensing issues; it might also prove useful for later maintenance activities.”

Probably not a substantial claim: may not pass S3 (load-bearing).

We additionally performed a manual verification of every subcode assigned to each task in Table 3, including those subcodes likely to influence code churn.

“We additionally performed a manual verification of every subcode assigned to each task in Table 3, including those subcodes likely to influence code churn.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

The limited analysis of current GenAI adoption policies represents an important research opportunity.

“The limited analysis of current GenAI adoption policies represents an important research opportunity.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

This reflects a growing recognition of the need for transparency in collaborative software development, where understanding the role of AI in contributions can improve accountability and ensure compliance.

“This reflects a growing recognition of the need for transparency in collaborative software development, where understanding the role of AI in contributions can improve accountability and ensure compliance.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The GitClear 2025 report reported a 39.9% drop in refactoring.

“a 39.9% drop in refactoring”

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

GenAI tools are actively used in open-source software and developers are working on guiding their usage.

“Our findings show that GenAI tools are actively used in open-source software and that developers are working on guiding their usage.”

Probably not a substantial claim: may not pass S4 (based).

Line-based churn captures the frequency with which individual lines are churned, indicating potential code maintainability challenges.

“This metric captures the frequency with which individual lines are churned, indicating potential code maintainability challenges.”

Probably not a substantial claim: may not pass S4 (based). May combine two claims.

GitHub Copilot is a tool tailored to software development and is more deeply embedded in developers' workflows than ChatGPT.

“For studies targeting GitHub repositories, it is crucial to consider GitHub Copilot as well, because (1) unlike ChatGPT, it is a tool tailored to software development, and (2) it is more deeply embedded in developers' workflows (their local editors, but also into the GitHub platform as a whole).”

Classified as probably not a substantial claim.

Code churn is particularly relevant for understanding the maintainability of LLM-generated source code, which might introduce redundancies or bugs that result in changes soon after adding generated code.

“Code churn is particularly relevant for understanding the maintainability of LLM-generated source code, which might introduce redundancies or bugs that result in changes soon after adding generated code.”

May combine two claims.

Across all robustness checks, the results remained stable, indicating that our main RDD findings do not depend on arbitrary cutoff choices or temporal window specifications.

“Across all robustness checks, the results remained stable, indicating that our main RDD findings do not depend on arbitrary cutoff choices or temporal window specifications.”

Probably not a substantial claim: may not pass S4 (based).

For downstream consumers of open-source dependencies, explicit GenAI policies serve as a signal of due diligence that may influence their dependency selection.

“For downstream consumers of opensource dependencies, explicit GenAI policies serve as a signal of due diligence that may influence their dependency selection.”

Probably not a substantial claim: may not pass S4 (based).

An important gap remains in our understanding of self-admitted GenAI usage patterns in the wild, particularly regarding how developers openly acknowledge and document GenAI assistance across different software engineering tasks and purposes.

“Despite the increasing amount of research studying GenAI assistance in software development, an important gap remains in our understanding of self-admitted GenAI usage patterns in the wild, particularly regarding how developers openly acknowledge and document GenAI assistance across different software engineering tasks and purposes.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The GitClear 2025 follow-up report, based on an expanded dataset of 211 million lines through 2024, reported a rise in code churn from 4.5% in 2023 to 5.7% in 2024.

“The 2025 follow-up report [52], based on an expanded dataset of 211 million lines through 2024, reported a rise in code churn from 4.5% in 2023 to 5.7% in 2024”

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing).

This response reflects the perspective that standard testing and code review are sufficient to ensure the quality of both AI-generated and manually created content.

“This response reflects the perspective that standard testing and code review are sufficient to ensure the quality of both AI-generated and manually created content.”

Probably not a substantial claim: may not pass S2 (checkable).

Human oversight is important in GenAI-assisted development.

“This, together with the human corrective actions triggered by GenAI actions we observed, points to the importance of human oversight in GenAI-assisted development.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Our results may not generalize to other open-source platforms such as GitLab, which may have different norms and adoption patterns.

“Moreover, our results may not generalize to other open-source platforms such as GitLab, which may have different norms and adoption patterns.”

Probably not a substantial claim: may not pass S3 (load-bearing).

We might have misclassified the scope and purpose of some GenAI mentions in some instances.

“Although we always examined the whole context around a GenAI mention, we might have misclassified its scope and purpose in some instances.”

May combine two claims.

Given our manual validation, the precision of the identified GenAI usages is high, even though the overall recall of all GenAI usages is low.

“Given our manual validation, the precision of the identified GenAI usages is high, even though the overall recall of all GenAI usages is low.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Our focus on ChatGPT and GitHub Copilot might not capture the usage patterns of other GenAI tools that were released more recently, e.g., Claude Code.

“Finally, our focus on ChatGPT and GitHub Copilot might not capture the usage patterns of other GenAI tools that were released more recently, e.g., Claude Code.”

Classified as probably not a substantial claim.

Our curated dataset of 1,292 self-admitted GenAI mentions and the taxonomy derived in RQ1 provide a foundation for scaling empirical studies via automated detection and large-scale mining.

“Our curated dataset of 1,292 self-admitted GenAI mentions and the taxonomy derived in RQ1 provide a foundation for scaling empirical studies via automated detection and large-scale mining.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The GitClear 2025 report reported that for the first time, copy-pasted lines exceeded moved lines within commits.

“reported that for the first time, copy-pasted lines exceeded moved lines within commits”

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S4 (based).

Unacknowledged usage before the first explicit mention could distort the pre-adoption baseline for our code churn analysis (RQ3).

“Finally, unacknowledged usage before the first explicit mention could distort the pre-adoption baseline for our code churn analysis (RQ3).”

Classified as probably not a substantial claim.

Our generous analysis window of 360 days before and after this point and robustness checks (e.g., placebo tests) help accommodate potential discrepancies in adoption dates.

“However, our generous analysis window of 360 days before and after this point and robustness checks (e.g., placebo tests) help accommodate potential discrepancies in adoption dates.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Security concerns with AI-generated code may reflect broader challenges in secure coding practices rather than AI-specific problems.

“These findings suggest that security concerns with AI-generated code may reflect broader challenges in secure coding practices rather than AI-specific problems.”

Probably not a substantial claim: may not pass S2 (checkable).

We consider self-admitted GenAI usage a useful analytical lens for studying the impact of GenAI adoption in open-source software projects.

“However, at the same time, we consider self-admitted GenAI usage a useful analytical lens for studying the impact of GenAI adoption in open-source software projects.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The self-admitted GenAI usage we studied might not reflect general GenAI usage-self-admitted or hidden-in other repositories, programming languages, or industrial software projects.

“The self-admitted GenAI usage we studied might not reflect general GenAI usage-self-admitted or hidden-in other repositories, programming languages, or industrial software projects.”

Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

Strict usage policies might lead to fewer self-admitted usages, as contributors might avoid disclosure (RQ2).

“Strict usage policies might lead to fewer self-admitted usages, as contributors might avoid disclosure (RQ2).”

Classified as probably not a substantial claim. May combine two claims.

The GitClear 2025 report indicated a fundamental shift away from code refactoring toward code duplication and raised concerns about growing technical debt and the long-term sustainability of AI-assisted coding.

“indicating a fundamental shift away from code refactoring toward code duplication and raising concerns about growing technical debt and the long-term sustainability of AI-assisted coding.”

• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Classified as probably not a substantial claim.

The observed frequencies might reflect what developers feel comfortable disclosing rather than their actual usage (RQ1).

“Therefore, the observed frequencies might reflect what developers feel comfortable disclosing rather than their actual usage (RQ1).”

Probably not a substantial claim: may not pass S2 (checkable).

Analyses that ignore project-specific adoption points or governance contexts risk drawing misleading conclusions.

“Analyses that ignore project-specific adoption points or governance contexts risk drawing misleading conclusions.”

May combine two claims.

The number of policies and guidelines we analyzed and the number of survey responses we received were relatively low.

“The number of policies and guidelines we analyzed and the number of survey responses we received were relatively low.”

Probably not a substantial claim: may not pass S3 (load-bearing).

We rely on manually annotated data, which may be miscoded due to the subjective nature of understanding the coding book.

“We rely on manually annotated data, which may be miscoded due to the subjective nature of understanding the coding book.”

May combine two claims.

To mitigate the miscoding threat and ensure consistency in our qualitative analysis, we implemented a rigorous manual review process with multiple raters in several rounds of independent coding, achieving high inter-rater reliability.

“To mitigate this threat and ensure consistency in our qualitative analysis, we implemented a rigorous manual review process with multiple raters in several rounds of independent coding, achieving high inter-rater reliability.”

Probably not a substantial claim: may not pass S4 (based).

Developers may be more likely to acknowledge GenAI usage for mundane tasks such as translation rather than for core development tasks such as implementing complex business logic.

“Developers may be more likely to acknowledge GenAI usage for mundane tasks such as translation rather than for core development tasks such as implementing complex business logic.”

May combine two claims.

Developers who use GenAI tools without leaving a trace remain outside of our analysis scope, meaning our findings represent a lower bound on actual GenAI adoption.

“Developers who use GenAI tools without leaving a trace remain outside of our analysis scope, meaning our findings represent a lower bound on actual GenAI adoption.”

Classified as probably not a substantial claim. May combine two claims.

We addressed the false negative threat by developing comprehensive regular expressions, covering common naming variations, and conducting a thorough manual validation of the identified mentions.

“We addressed this by developing comprehensive regular expressions, covering common naming variations, and conducting a thorough manual validation of the identified mentions.”