From Human Oversight to Cognitive Sovereignty: A Process-Based Governance Standard for AI-Assisted Legal Reasoning

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

Smuha (2021) treats the risk-based model as substantially system-centric.

“Smuha (2021) treats the risk-based model as substantially system-centric.”

• cites Smuha (2021) sourceDisputed

  Section 2.2 (‘Regulating AI—Not a walk in the park’; page unavailable) makes the contrary point: requirements for trustworthy AI concern human behaviour rather than AI systems themselves, and must account for actors, processes and the wider context of use. The paper does not describe its approach as a ‘risk-based model,’ but its stated framing is not substantially system-centric. (LLM, conf=0.76)

  Checked by a reasoning model — may be wrong.

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim.

Mittelstadt (2019) argues that high-level AI principles cannot guarantee ethical AI because they lack institutional translation, professional norms and robust accountability mechanisms.

“Mittelstadt (2019) argues that high-level AI principles cannot guarantee ethical AI because they lack institutional translation, professional norms and robust accountability mechanisms.”

• cites Mittelstadt (2019) sourceSupported

Classified as probably not a substantial claim.

Veale and Zuiderveen Borgesius (2021) argue that conformity-assessment architecture can overemphasise provider-side obligations while leaving downstream use practices under-specified.

“Veale and Zuiderveen Borgesius (2021) argue that conformityassessment architecture can overemphasise providerside obligations while leaving downstream use practices under-specified.”

• cites Veale and Zuiderveen Borgesius (2021) sourceSupported

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

The concern is not only whether AI improves efficiency but also whether algorithmic assistance remains compatible with accountable justification, discretionary control, and rule-of-law legitimacy.

“In public-law terms, the concern is not only whether AI improves efficiency but also whether algorithmic assistance remains compatible with accountable justification, discretionary control, and rule-of-law legitimacy (Danaher, 2016;Binns, 2018;Busuioc, 2021).”

• cites Danaher, 2016 sourceSupported
• cites Binns, 2018 sourcePartly supported

  The Introduction and Conclusion argue that algorithmic decisions require accountable justification, while the section on substantive laws connects that accountability to legal constraints. The cited passages do not address preserving discretionary control over algorithmic assistance, so Binns supports only part of the attributed concern. The indexed passages provide section labels but no page numbers. (LLM, conf=0.83)

  Experimental — accuracy not yet measured
• cites Busuioc, 2021 sourceSupported

Classified as probably not a substantial claim.

Lazcoz and de Hert (2023) warn that human intervention can become a mechanism for responsibility absorption when the human is present formally but does not exercise meaningful control.

“Lazcoz and de Hert (2023) warn that human intervention can become a mechanism for responsibility absorption when the human is present formally but does not exercise meaningful control.”

• cites Lazcoz and de Hert (2023) sourcePartly supported

  Section 7 warns that adding a human to the loop can let controllers avoid safeguards, while insisting that intervention must be meaningful rather than token. Section 12 describes human agents who routinely apply algorithmic outputs without improving decisions. These passages support the warning about merely formal human involvement, but the paper does not describe that involvement as a mechanism for absorbing responsibility; Section 9 places accountability with controllers. (LLM, conf=0.81)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim.

Cobbe (2019) extends the point to judicial review, showing that courts need legally intelligible criteria for examining automated public-sector decisionmaking.

“Cobbe (2019) extends the point to judicial review, showing that courts need legally intelligible criteria for examining automated public-sector decisionmaking.”

• cites Cobbe (2019) sourceSupported

Classified as probably not a substantial claim.

Cognitive Sovereignty responds to the threat of algocracy: the risk that governance becomes dependent on decision systems whose operation is not meaningfully contestable by the persons subject to them.

“In this sense, Cognitive Sovereignty responds to the threat of algocracy: the risk that governance becomes dependent on decision systems whose operation is not meaningfully contestable by the persons subject to them (Danaher, 2016).”

• cites Danaher, 2016 sourceSupported

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

Cognitive Sovereignty reflects the public-reason requirement that exercises of authority be justified through reasons that can be inspected, contested and institutionally attributed.

“It also reflects the public-reason requirement that exercises of authority be justified through reasons that can be inspected, contested and institutionally attributed (Binns, 2018).”

• cites Binns, 2018 sourceSupported

Classified as probably not a substantial claim.

Binns (2018) frames algorithmic accountability as a matter of public reason: affected persons and institutions require intelligible justifications, not merely correct outputs.

“Binns (2018) frames algorithmic accountability as a matter of public reason: affected persons and institutions require intelligible justifications, not merely correct outputs.”

• cites Binns (2018) sourceSupported

Classified as probably not a substantial claim.

Busuioc (2021) argues that AI systems generate distinct accountability challenges because responsibility can become distributed across technical systems, organisations and human officials.

“Busuioc (2021) argues that AI systems generate distinct accountability challenges because responsibility can become distributed across technical systems, organisations and human officials.”

• cites Busuioc (2021) sourceSupported

Classified as probably not a substantial claim.

Gummadi Usha Rani illustrates the governance problem on the judicial side.

“Gummadi Usha Rani illustrates the problem on the judicial side.”

• cites Gummadi Usha Rani v. Sure Mallikarjuna RaoNot matched

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

May combine two claims.

In Mata v. Avianca, the researching attorney used ChatGPT as his sole research source, attempted to locate at least one of the generated cases, failed to find it and nevertheless allowed the submission to proceed.

“The researching attorney used ChatGPT as his sole research source, attempted to locate at least one of the generated cases, failed to find it and nevertheless allowed the submission to proceed.”

• cites Mata v. Avianca, Inc.Not matched

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

Classified as probably not a substantial claim.

ABA Formal Opinion 512 warns that uncritical reliance, without appropriate independent verification, may breach the duty of competence.

“and warns that uncritical reliance, without appropriate independent verification, may breach the duty of competence.”

• cites American Bar Association's Formal Opinion 512 (2024)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.

A false precedent inserted into a judgment or pleading affects more than the lawyer's competence; it threatens the legitimacy of the institutional process through which law is publicly justified.

“A false precedent inserted into a judgment or pleading therefore affects more than the lawyer's competence.It threatens the legitimacy of the institutional process through which law is publicly justified.”

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

AI-assisted legal reasoning should consequently be analysed as a problem of governance, not only as a problem of ethics or legal technology.

“AI-assisted legal reasoning should consequently be analysed as a problem of governance, not only as a problem of ethics or legal technology.”

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

In that setting, AI does not merely supply information; it can structure the professional's cognition.

“In that setting, AI does not merely supply information; it can structure the professional's cognition.”

Classified as probably not a substantial claim.

The arguments from AI ethics and explainability literature reinforce the need for SIM and ART.

“These arguments reinforce the need for SIM and ART.”

Classified as probably not a substantial claim.

Wachter, Mittelstadt and Russell (2021) make the parallel point in EU nondiscrimination law: legal standards often require contextual and interpretive judgment that cannot be reduced to static technical metrics.

“Wachter, Mittelstadt and Russell (2021) make the parallel point in EU nondiscrimination law: legal standards often require contextual and interpretive judgment that cannot be reduced to static technical metrics.”

• cites Wachter, Mittelstadt and Russell (2021) sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

Fink (2025) argues that Article 14's effectiveness depends on implementation that accounts for cognitive constraints and automation bias.

“Fink (2025) similarly argues that Article 14's effectiveness depends on implementation that accounts for cognitive constraints and automation bias.”

• cites Fink (2025)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).

Yeung (2018) conceptualises algorithmic regulation as a mode of governance that uses computational systems to monitor, evaluate and shape behaviour.

“Yeung (2018) conceptualises algorithmic regulation as a mode of governance that uses computational systems to monitor, evaluate and shape behaviour.”

• cites Yeung (2018) sourceNot checked — no full text of the source

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

Alarie, Niblett and Yoon (2018) show that AI is already reshaping the organisation and delivery of legal services.

“Alarie, Niblett and Yoon (2018) show that AI is already reshaping the organisation and delivery of legal services”

• cites Alarie, Niblett and Yoon (2018) sourceNot checked — no full text of the source

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

Generative AI has entered legal work faster than the legal profession has developed institutional rules for its responsible use.

“Generative artificial intelligence has entered legal work faster than the legal profession has developed institutional rules for its responsible use.”

Classified as probably not a substantial claim.

ABA Formal Opinion 512 requires a reasonable understanding of the specific tool's capabilities and limitations.

“It requires a reasonable understanding of the specific tool's capabilities and limitations”

• cites American Bar Association's Formal Opinion 512 (2024)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).

Human oversight must be operationalised as a reviewable process rather than left as a mere formal label.

“These works support the article's claim that human oversight must be operationalised as a reviewable process rather than left as a mere formal label.”

Classified as probably not a substantial claim.

Zalnieriute, Moses, and Williams (2019) show that automation in government decisionmaking creates rule-of-law tensions in which legal decisions affecting rights and entitlements become difficult to contest or explain.

“Zalnieriute, Moses, and Williams (2019) show that automation in government decisionmaking creates rule-of-law tensions in which legal decisions affecting rights and entitlements become difficult to contest or explain.”

• cites Zalnieriute, Moses, and Williams (2019) sourceNot checked — no full text of the source

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

The Supreme Court of India stated that reliance on nonexistent and fake alleged judgments was not merely an error in decision-making but misconduct with legal consequences.

“It stated that reliance on nonexistent and fake alleged judgments was not merely an error in decision-making but misconduct with legal consequences (Gummadi Usha Rani, 2026, para. 7).”

• cites Gummadi Usha Rani, 2026Not matched

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

May combine two claims.

The most visible failures have involved fabricated authorities, but the deeper problem is not confined to hallucinated citations.

“The most visible failures have involved fabricated authorities, but the deeper problem is not confined to hallucinated citations.”

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

Chesterman (2021) identifies opacity as a central regulatory problem in AI systems.

“Chesterman (2021) identifies opacity as a central regulatory problem in AI systems.”

• cites Chesterman (2021) sourceNot checked — no full text of the source

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

Oswald (2018) argues that algorithm-assisted public-sector decisionmaking must be assessed through administrative-law concepts such as discretion, procedural fairness and reviewability.

“Oswald (2018) similarly argues that algorithm-assisted public-sector decisionmaking must be assessed through administrative-law concepts such as discretion, procedural fairness and reviewability.”

• cites Oswald (2018) sourceNot checked — no full text of the source

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

Legal reasoning shapes the exercise of coercive public power, the allocation of rights and duties, and public trust in courts and regulatory institutions.

“It shapes the exercise of coercive public power, the allocation of rights and duties, and public trust in courts and regulatory institutions.”

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

SIM and ART can be implemented without waiting for new legislation, because they can operate through judicial practice directions, bar guidance, procurement conditions, professional training and internal compliance rules.

“Both can be implemented without waiting for new legislation, because they can operate through judicial practice directions, bar guidance, procurement conditions, professional training and internal compliance rules.”

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

A lawyer may certify that authorities have been checked while relying on AI-generated summaries, or may read a fluent AI draft for style while leaving its doctrinal structure intact.

“A lawyer may certify that authorities have been checked while relying on AI-generated summaries, or may read a fluent AI draft for style while leaving its doctrinal structure intact.”

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

Competence assesses whether the final work product satisfies professional obligations.

“Competence assesses whether the final work product satisfies professional obligations.”

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

The framework is transferable beyond law, but transfer should be cautious.

“The framework is also transferable beyond law, but transfer should be cautious.”

May combine two claims.

In Mata, the initial prompt effectively asked ChatGPT to produce support for a desired tolling argument; the legal viability of that argument was not first established through independent legal analysis.

“In Mata, the initial prompt effectively asked ChatGPT to produce support for a desired tolling argument; the legal viability of that argument was not first established through independent legal analysis.”

• cites Mata v. Avianca, Inc. (678 F. Supp.3d 443, S.D.N.Y. 2023)Not matched

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

Classified as probably not a substantial claim.

Stage 4 is the evidentiary gate through which AI-generated material must pass before it can be used in legal reasoning.

“It is the evidentiary gate through which AI-generated material must pass before it can be used in legal reasoning.”

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

Generative AI failures in legal practice are often described as hallucination problems, but that description is incomplete.

“Generative AI failures in legal practice are often described as hallucination problems.”

• Also stated in: § 9. Conclusions and Policy Implications · plain

  “That description is incomplete.”

• AssumptionRests on an assumption: Generative AI failures in legal practice are often described as hallucination problems.

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

Explanation alone is not enough.

“This is also why explanation alone is not enough.”

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

The EU AI Act's value here is distinct: it is the most developed legislative example of human oversight as a requirement of AI governance.

“Its value here is distinct: it is the most developed legislative example of human oversight as a requirement of AI governance.”

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

Stage 3 directly counters System 1 Lawyering by forcing the professional to engage System 2.

“This stage directly counters System 1 Lawyering by forcing the professional to engage System 2.”

Classified as probably not a substantial claim.

The conventional inquiry asks whether the final document contains a false citation or a misleading proposition.

“The conventional inquiry asks whether the final document contains a false citation or a misleading proposition.”

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

The same general sequence - frame, expand, reconstruct, verify and judge - may be useful in medicine, financial auditing and public administration.

“The same general sequence -frame, expand, reconstruct, verify and judge -may be useful in those domains.”

May combine two claims.

AI tools are likely to affect legal research, prediction, drafting, billing models and access to justice, thereby changing the ordinary infrastructure of legal service delivery.

“AI tools are likely to affect legal research, prediction, drafting, billing models and access to justice, thereby changing the ordinary infrastructure of legal service delivery (Alarie et al., 2018).”

• cites Alarie et al., 2018 sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

ART translates accountability concerns into the legal-professional context by requiring a record of framing, supervision, verification and independent judgment.

“ART translates those concerns into the legal-professional context by requiring a record of framing, supervision, verification and independent judgment.”

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

For Central and Eastern European regulatory contexts, the framework is especially practical.

“For Central and Eastern European regulatory contexts, the framework is especially practical.”

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

Medicine, financial auditing and public administration all involve professional reliance on AI-generated recommendations.

“Medicine, financial auditing and public administration all involve professional reliance on AI-generated recommendations.”

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

Current legal and regulatory frameworks increasingly require a human presence; they less often define the cognitive operations that make that presence meaningful.

“Current legal and regulatory frameworks increasingly require a human presence; they less often define the cognitive operations that make that presence meaningful.”

May combine two claims.

Explainability may assist governance, but explanation does not itself prove that independent legal judgment occurred.

“and explainability may assist governance, but explanation does not itself prove that independent legal judgment occurred.”

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

ART asks how the document came to contain a false citation or misleading proposition, and at which point the reasoning chain failed.

“ART asks how the document came to contain it, and at which point the reasoning chain failed.”

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

In England and Wales, the Solicitors Regulation Authority's risk outlook on AI emphasises that firms remain responsible for outcomes, that AI systems may generate inaccurate material and that firms should not trust an AI system to judge its own accuracy.

“In England and Wales, the Solicitors Regulation Authority's risk outlook on AI emphasises that firms remain responsible for outcomes, that AI systems may generate inaccurate material and that firms should not trust an AI system to judge its own accuracy.”

• cites Solicitors Regulation Authority's risk outlook on AINot matched

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

May combine two claims.

Parasuraman and Riley (1997) distinguished automation use, misuse, disuse and abuse, showing that human reliance on automated systems is not a simple matter of rational tool adoption.

“Parasuraman and Riley (1997) distinguished automation use, misuse, disuse and abuse, showing that human reliance on automated systems is not a simple matter of rational tool adoption.”

• cites Parasuraman and Riley (1997) sourceNot checked — no full text of the source

May combine two claims.

Professional conduct rules identify breaches only after the failure has occurred; they rarely provide the procedural architecture to prevent the failure.

“Professional conduct rules identify breaches only after the failure has occurred; they rarely provide the procedural architecture to prevent the failure.”

Classified as probably not a substantial claim.

ART shifts institutional review from a narrow outcome question to a process question.

“ART shifts institutional review from a narrow outcome question to a process question.”

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

The transformation of legal work by AI makes general competence duties necessary but insufficient: the more AI becomes embedded in ordinary legal work, the more professional responsibility needs a concrete process standard for reliance, verification and judgment.

“That transformation makes general competence duties necessary but insufficient: the more AI becomes embedded in ordinary legal work, the more professional responsibility needs a concrete process standard for reliance, verification and judgment.”

Classified as probably not a substantial claim.

Stage 1, Legal Framing, prevents algorithmic anchoring.

“This prevents algorithmic anchoring.”

Classified as probably not a substantial claim.

Outcome standards identify violations after they become visible.

“Outcome standards identify violations after they become visible.”

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

Professional conduct rules govern competence, candour, diligence and confidentiality.

“Professional conduct rules govern competence, candour, diligence and confidentiality.”

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

The missing object is not another warning that hallucinations are dangerous; it is a reviewable standard for how a legal professional must move from AI output to legally accountable judgment.

“The missing object is not another warning that hallucinations are dangerous.It is a reviewable standard for how a legal professional must move from AI output to legally accountable judgment.”

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

Editing preserves the machine's argumentative architecture; reconstruction makes the AI output only raw material.

“Editing preserves the machine's argumentative architecture; reconstruction makes the AI output only raw material.”

• AssumptionRests on an assumption: Editing an AI draft preserves the machine's argumentative architecture, while reconstruction makes the AI output only raw material.

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

Risk-based AI regulation governs systems, providers' duties, and deployers' obligations.

“Risk-based AI regulation governs systems, providers' duties, and deployers' obligations.”

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

Lee and See (2004) argue that appropriate reliance depends on calibrated trust: users may misuse automation when trust exceeds system capability and disuse it when trust falls below capability.

“Lee and See (2004) argue that appropriate reliance depends on calibrated trust: users may misuse automation when trust exceeds system capability and disuse it when trust falls below capability.”

• cites Lee and See (2004) sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

The EU AI Act is not complete code for all AI use by lawyers.

“It is not complete code for all AI use by lawyers”

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

Legal reasoning has distinctive rule-of-law stakes: it authorises public coercion and must be justified through traceable sources.

“However, legal reasoning has distinctive rule-of-law stakes: it authorises public coercion and must be justified through traceable sources.”

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

In legal reasoning, AI assistance is legitimate only when the professional can reconstruct and defend the legal pathway independently of the system that suggested it.

“In legal reasoning, this means that AI assistance is legitimate only when the professional can reconstruct and defend the legal pathway independently of the system that suggested it.”

May combine two claims.

Endsley (2017) adds that highly automated environments create out-of-the-loop performance problems, weakening the human operator's situation awareness and ability to intervene.

“Endsley (2017) adds that highly automated environments create out-of-the-loop performance problems, weakening the human operator's situation awareness and ability to intervene.”

• cites Endsley (2017) sourceNot checked — no full text of the source

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

Where Stage 4 verification is not satisfied, the proper conclusion is that a prima facie procedural breach exists.

“Where Stage 4 verification is not satisfied, the proper conclusion is that a prima facie procedural breach exists.”

May combine two claims.

SIM and ART do not regulate model design, training data, cybersecurity, privacy or provider liability.

“First, SIM and ART do not regulate model design, training data, cybersecurity, privacy or provider liability.”

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

Intent, recklessness, negligence, and good faith remain relevant to sanctions, remedies, and the severity of disciplinary action.

“Intent, recklessness, negligence, and good faith remain relevant to sanctions, remedies, and the severity of disciplinary action.”

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

In such cases, professional responsibility exists in the abstract but lacks an operational test for whether independent legal cognition occurred.

“In such cases, professional responsibility exists in the abstract but lacks an operational test for whether independent legal cognition occurred.”

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

Law is the appropriate domain in which to formulate the doctrine first.

“That is why law is the appropriate domain in which to formulate the doctrine first.”

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

The literature converges on a precise gap.

“The literature converges on a precise gap.”

May combine two claims.

Both Gummadi Usha Rani and Mata would have failed at Stage 4.

“Both Gummadi Usha Rani and Mata would have failed at this point.”

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

A negative answer to any ART step identifies the corresponding SIM stage and enables targeted findings: failure of framing, supervision, verification, or judgment.

“A negative answer to any step identifies the corresponding SIM stage and enables targeted findings: failure of framing, supervision, verification, or judgment.”

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

Verification duties require that material be accurate.

“Verification duties require that material be accurate.”

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

Member States and professional bodies will need guidance, training protocols, and audit standards for the use of AI in courts, legal aid, public administration, and regulated professions.

“Member States and professional bodies will need guidance, training protocols, and audit standards for the use of AI in courts, legal aid, public administration, and regulated professions.”

Classified as probably not a substantial claim.

The EU AI Act's weakness for present purposes is that it does not translate oversight into a professional reasoning protocol.

“The Act's weakness for present purposes is that it does not translate oversight into a professional reasoning protocol.”

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

AI is a search accelerator at Stage 2, not a legal reasoner whose outputs carry authority.

“AI is a search accelerator at this stage, not a legal reasoner whose outputs carry authority.”

• AssumptionRests on an assumption: AI is a search accelerator, not a legal reasoner whose outputs carry authority.

Classified as probably not a substantial claim.

Stage 4 is not ordinary proofreading.

“This stage is not ordinary proofreading.”

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

The key distinction in Stage 3 is between editing an AI draft and rebuilding the legal analysis.

“The key distinction is between editing an AI draft and rebuilding the legal analysis.”

Classified as probably not a substantial claim.

SIM and ART address the professional reasoning layer.

“They address the professional reasoning layer.”

May combine two claims.

Existing frameworks do important work: professional conduct rules impose duties of competence, diligence and candour; ethical guidance alerts lawyers to hallucination and confidentiality risks; and the EU AI Act creates system-level human-oversight duties for high-risk AI systems.

“Existing frameworks do important work: professional conduct rules impose duties of competence, diligence and candour; ethical guidance alerts lawyers to hallucination and confidentiality risks; and the EU Artificial Intelligence Act creates system-level humanoversight duties for high-risk AI systems.”

• cites Regulation (EU) 2024/1689Not 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).

The five operational elements of Cognitive Sovereignty are not optional virtues but the minimum conditions under which human oversight can be more than symbolic.

“These are not optional virtues.They are the minimum conditions under which human oversight can be more than symbolic.”

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

Without the distinction between supervised intelligence and delegated judgment, human oversight risks becoming a ritual of responsibility without actual cognitive control.

“Without that distinction, human oversight risks becoming a ritual of responsibility without actual cognitive control.”

May combine two claims.

Conduct-based and system-based governance are increasingly developed, while process-based governance of the cognitive operations connecting AI output to professional judgment remains underspecified.

“The first two levels are increasingly developed.The third remains underspecified.”

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

All substantive claims, legal interpretations, case references, citations, references, arguments and final wording were reviewed, verified and approved by the author before submission.

“All substantive claims, legal interpretations, case references, citations, references, arguments and final wording were reviewed, verified and approved by the author before submission.”

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

A judge who signs an AI-drafted order without reconstructing the authorities is in the loop formally but not cognitively.

“A judge who signs an AI-drafted order without reconstructing the authorities is in the loop formally but not cognitively.”

May combine two claims.

Responsibility appears to remain with the human actor, but the actual structure of reasoning has been partly delegated to a system that cannot bear legal or democratic responsibility.

“Responsibility appears to remain with the human actor, but the actual structure of reasoning has been partly delegated to a system that cannot bear legal or democratic responsibility.”

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

Ethical review and approval were not required for this study because it did not involve human participants, interviews, surveys, experiments, recordings, classroom data, institutional data, personal data, vulnerable groups or sensitive material.

“Ethical review and approval were not required for this study because it did not involve human participants, interviews, surveys, experiments, recordings, classroom data, institutional data, personal data, vulnerable groups or sensitive material.”

Classified as probably not a substantial claim.

A human can be in the loop while performing no meaningful cognitive function.

“A human can be in the loop while performing no meaningful cognitive function.”

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

Gummadi Usha Rani and Mata v. Avianca are diagnostic cases, not a representative sample of all AI failures in legal practice.

“Gummadi Usha Rani and Mata v. Avianca are diagnostic cases, not a representative sample of all AI failures in legal practice.”

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

Cognitive Sovereignty requires that the human actor remain the source of legal authority within the reasoning process.

“Cognitive Sovereignty requires that the human actor remain the source of legal authority within the reasoning process.”

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

The dominant public-law model of AI governance is risk-based regulation.

“The dominant public-law model of AI governance is risk-based regulation.”

• cites Regulation (EU) 2024/1689Not 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), S4 (based).

With that distinction, AI can augment legal work without displacing the human authority on which legal accountability and democratic legitimacy depend.

“With it, AI can augment legal work without displacing the human authority on which legal accountability and democratic legitimacy depend.”

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

The author used ChatGPT to support language editing, stylistic refinement, Polish translation assistance, structural polishing, APA-style formatting, reference list preparation and editorial revision suggestions, and the tool was not treated as an authoritative source of law, policy, evidence, data or bibliography.

“The author used ChatGPT (OpenAI) to support language editing, stylistic refinement, Polish translation assistance, structural polishing, APA-style formatting, reference list preparation, and the preparation of editorial revision suggestions.The tool was not treated as an authoritative source of law, policy, evidence, data or bibliography.”

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

The EU AI Act is used as a governance benchmark for human oversight, not as proof that every legal use of a general-purpose model is directly regulated as high-risk.

“The Act is used as a governance benchmark for human oversight, not as proof that every legal use of a general-purpose model is directly regulated as high-risk.”

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

Cognitive Sovereignty differs from a generic duty to remain in the loop.

“Cognitive Sovereignty differs from a generic duty to remain “in the loop”.”

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

The article uses two documented legal failures, Gummadi Usha Rani and Mata v. Avianca, as diagnostic case studies, selected because each exposes the same cognitive failure in different institutional settings.

“It uses two documented legal failures, Gummadi Usha Rani and Mata v. Avianca, as diagnostic case studies.They are not presented as a comprehensive dataset; they are selected because each exposes the same cognitive failure in different institutional settings: a judicial order in India and a lawyer-filed submission in the United States.”

May combine two claims.

System 1 processing is fast, automatic and pattern-driven, while System 2 processing is slower, effortful and analytic.

“System 1 processing is fast, automatic and pattern-driven; System 2 processing is slower, effortful and analytic.”

• cites Kahneman, D. (2011)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).

Courts and legal professionals participate in the production of public reasons.

“Courts and legal professionals participate in the production of public reasons.”

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

The article does not claim that all legal uses of AI fall within the high-risk categories of the EU AI Act.

“Second, the article does not claim that all legal uses of AI fall within the high-risk categories of the EU AI Act.”

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

Generative AI produces material with the surface form of professional authority: plausible case names, citation formats, statutory references and coherent argumentative structure.

“Generative AI produces material with the surface form of professional authority: plausible case names, citation formats, statutory references and coherent argumentative structure.”

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

No new empirical data were generated or analysed in this study; the article is based on conceptual and doctrinal analysis of publicly available scholarly literature, legal materials, policy documents, regulatory instruments and judicial decisions.

“No new empirical data were generated or analysed during this study.The article is based on conceptual and doctrinal analysis of publicly available scholarly literature, legal materials, policy documents, regulatory instruments and judicial decisions cited in the References section.”

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

The proposed framework is not claimed to be positive law; it is proposed as a minimum procedural architecture that courts, regulators, bar associations and legal organisations can adopt or adapt.

“is not claimed to be positive law.It is proposed as a minimum procedural architecture that courts, regulators, bar associations and legal organisations can adopt or adapt.”

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

Where generative AI is used in legal reasoning, institutions need a process-level standard to preserve human cognitive authority and to audit reliance.

“The more defensible claim is narrower: where generative AI is used in legal reasoning, institutions need a process-level standard to preserve human cognitive authority and to audit reliance.”

May combine two claims.

Trust calibration, automation bias, and out-of-the-loop performance problems are persistent human factors risks.

“Trust calibration, automation bias, and out-of-the-loop performance problems are persistent human factors risks (Lee & See, 2004;Hoff & Bashir, 2015;Endsley, 2017).”

• cites Lee & See, 2004 sourceNot checked — no full text of the source
• cites Hoff & Bashir, 2015 sourceNot checked — no full text of the source
• cites Endsley, 2017 sourceNot checked — no full text of the source

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

The framework does not claim to eliminate all forms of inappropriate reliance.

“A further limitation is that the framework does not claim to eliminate all forms of inappropriate reliance.”

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

The proposed framework is not yet empirically validated.

“Third, the framework is not yet empirically validated.”

May combine two claims.

Article 14 of the EU AI Act requires that overseers of high-risk AI systems be enabled to understand system capacities and limitations, remain aware of automation bias, correctly interpret outputs, disregard or override outputs where necessary, and intervene in system operation.

“It also requires that overseers be enabled to understand system capacities and limitations, remain aware of automation bias, correctly interpret outputs, disregard or override outputs where necessary, and intervene in system operation.”

• cites Regulation (EU) 2024/1689Not matched

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

Classified as probably not a substantial claim.

Cognitive Sovereignty addresses the legitimacy deficit by insisting that the human actor's role be substantively auditable.

“Cognitive Sovereignty addresses that deficit by insisting that the human actor's role be substantively auditable.”

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

ART is not an automatic disciplinary machine; it identifies procedural failure, and institutions must still assess causation, harm, intent, proportionality and appropriate sanction.

“Fourth, ART is not an automatic disciplinary machine.It identifies procedural failure; institutions must still assess causation, harm, intent, proportionality and appropriate sanction.”

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

Cognitive Sovereignty specifies what the competent person must do when oversight concerns legal reasoning: frame independently, evaluate AI output critically, reconstruct from sources, verify and decide.

“Cognitive Sovereignty specifies what the competent person must do when oversight concerns legal reasoning: frame independently, evaluate AI output critically, reconstruct from sources, verify and decide.”

May combine two claims.

Existing governance frameworks (professional conduct rules, ethical guidance and the EU AI Act) do not sufficiently specify the cognitive process through which a professional must integrate AI output into legal reasoning, and do not provide institutions with a reviewable standard for the reasoning sequence itself.

“The gap is that these frameworks do not sufficiently specify the cognitive process through which a professional must integrate AI output into legal reasoning.”

• Also stated in: § 9. Conclusions and Policy Implications · plain

  “They regulate systems, conduct, and outcomes; they do not yet provide institutions with a reviewable standard for the reasoning sequence itself.”

The quote only partly states this claim.

The governance failure can be stated as a three-level mismatch between conduct-based, system-based and process-based governance.

“The governance failure can be stated as a three-level mismatch.”

May combine two claims.

The EU Artificial Intelligence Act, Regulation (EU) 2024/1689, classifies AI systems according to risk and imposes graduated duties on providers and deployers.

“The EU Artificial Intelligence Act, Regulation (EU) 2024/1689, 2024, classifies AI systems according to risk and imposes graduated duties on providers and deployers.”

• cites Regulation (EU) 2024/1689Not matched

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