pith. machine review for the scientific record. sign in

arxiv: 2604.04604 · v1 · submitted 2026-04-06 · 💻 cs.CY · cs.AI· cs.CR· cs.MA

Recognition: 2 theorem links

· Lean Theorem

AI Agents Under EU Law

Luca Nannini , Adam Leon Smith , Michele Joshua Maggini , Enrico Panai , Sandra Feliciano , Aleksandr Tiulkanov , Elena Maran , James Gealy , Piercosma Bisconti

Authors on Pith no claims yet

Pith reviewed 2026-05-10 19:36 UTC · model grok-4.3

classification 💻 cs.CY cs.AIcs.CRcs.MA
keywords AI agentsEU AI Actregulatory compliancebehavioral drifthigh-risk systemscompliance architectureagent taxonomymulti-party action chains
0
0 comments X

The pith

High-risk AI agents with untraceable behavioral drift cannot currently satisfy the EU AI Act's essential requirements.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper maps obligations for AI agents under the EU AI Act alongside GDPR, the Cyber Resilience Act, and other rules. It supplies a taxonomy of nine deployment categories and a twelve-step compliance architecture to connect specific agent actions to regulatory triggers. The core finding is that providers must perform an exhaustive inventory of external actions, data flows, connected systems, and affected persons. Without traceable behavior, high-risk agents cannot meet the Act's essential requirements even after applying current draft standards. This matters because agents are already scaling into customer service, clinical support, and infrastructure where non-compliance blocks deployment.

Core claim

The authors argue that high-risk agentic systems with untraceable behavioral drift cannot currently satisfy the AI Act's essential requirements. The provider's foundational compliance task is an exhaustive inventory of the agent's external actions, data flows, connected systems, and affected persons. This conclusion rests on integration of draft harmonised standards under M/613, the GPAI Code of Practice, CRA standards under M/606, and Digital Omnibus proposals, supported by a taxonomy of nine agent deployment categories and a regulatory trigger mapping that links concrete actions to applicable legislation.

What carries the argument

The twelve-step compliance architecture and the taxonomy of nine agent deployment categories that map concrete actions to regulatory triggers across the AI Act and overlapping EU laws.

If this is right

  • Providers must inventory every external action, data flow, and affected person before deploying high-risk agents.
  • Untraceable drift creates a structural barrier that current standards cannot overcome for high-risk systems.
  • Compliance architecture must simultaneously address the AI Act, GDPR, Cyber Resilience Act, and sector rules.
  • The regulatory trigger mapping allows specific agent behaviors to be tied directly to required obligations.
  • Failure to complete the inventory leaves high-risk agents ineligible for EU market placement.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Agent designs will need built-in audit logs and observability features to enable the required inventory.
  • The inventory approach could inform compliance strategies in other jurisdictions adopting similar risk-based AI rules.
  • Developers may need to restrict autonomy levels in high-risk domains until traceability improves.
  • Standardised logging tools for multi-party action chains could emerge as a practical next step for industry.

Load-bearing premise

The assumption that the draft harmonised standards under M/613, the GPAI Code of Practice, CRA standards under M/606, and Digital Omnibus proposals will accurately represent the final enforceable obligations.

What would settle it

A documented high-risk AI agent that exhibits untraceable behavioral drift yet passes an official AI Act conformity assessment after the twelve-step compliance process is applied.

Figures

Figures reproduced from arXiv: 2604.04604 by Adam Leon Smith, Aleksandr Tiulkanov, Elena Maran, Enrico Panai, James Gealy, Luca Nannini, Michele Joshua Maggini, Piercosma Bisconti, Sandra Feliciano.

Figure 1
Figure 1. Figure 1: Non-exhaustive taxonomy of AI Agent Use Cases and Actions, detailing the concrete tasks performed across different domains using a shared LLM-based architecture. For economic operators building general-purpose agent platforms (i.e., platforms designed for arbitrary deployment by downstream deployers), this creates a classification dilemma. If the platform cannot predict at design time how deployers will us… view at source ↗
Figure 2
Figure 2. Figure 2: Multi-layer regulatory architecture for AI agent providers. The agent’s external actions (top) determine which obligations are activated. The provider’s primary compliance scope (top three layers) flows through the AI Act essential requirements, the harmonised standards, and the GPAI model layer. Adjacent EU instruments (bottom six boxes) are triggered by the agent’s external effects: which data is process… view at source ↗
Figure 3
Figure 3. Figure 3: Operational mapping of agent-specific characteristics to amplified compliance challenges and the structural solutions required by the AI Act’s essential requirements as operationalised through the harmonised standards discussed in Section 6. The image visually contrasts ’Agent Characteristic’ with ’Amplified Compliance Challenge,’ leading to the specific ’Operational Solution.’ The diagram reinforces that … view at source ↗
Figure 4
Figure 4. Figure 4: The Multi-Layer Compliance Architecture for AI under EU Law. The diagram illustrates that the AI Act is only one layer in a complex, intersecting regulatory ecosystem, where horizontal frameworks (GDPR, Data Act, CRA) and sectoral rules must be applied simultaneously based on the agent’s specific context. 7.1 The New Legislative Framework as Interpretive Infrastructure The AI Act is not a standalone instru… view at source ↗
Figure 5
Figure 5. Figure 5: Twelve-step compliance sequence for AI agent providers (Section 8.1). Three decision nodes determine the compliance pathway: Step 0 gates AI Act applicability (non-AI systems face only adjacent legislation); Step 2 determines whether full Chapter III essential requirements or Article 50 transparency obligations apply; Step 8 determines whether CRA product cybersecurity runs in parallel. The dotted feedback… view at source ↗
read the original abstract

AI agents - i.e. AI systems that autonomously plan, invoke external tools, and execute multi-step action chains with reduced human involvement - are being deployed at scale across enterprise functions ranging from customer service and recruitment to clinical decision support and critical infrastructure management. The EU AI Act (Regulation 2024/1689) regulates these systems through a risk-based framework, but it does not operate in isolation: providers face simultaneous obligations under the GDPR, the Cyber Resilience Act, the Digital Services Act, the Data Act, the Data Governance Act, sector-specific legislation, the NIS2 Directive, and the revised Product Liability Directive. This paper provides the first systematic regulatory mapping for AI agent providers integrating (a) draft harmonised standards under Standardisation Request M/613 to CEN/CENELEC JTC 21 as of January 2026, (b) the GPAI Code of Practice published in July 2025, (c) the CRA harmonised standards programme under Mandate M/606 accepted in April 2025, and (d) the Digital Omnibus proposals of November 2025. We present a practical taxonomy of nine agent deployment categories mapping concrete actions to regulatory triggers, identify agent-specific compliance challenges in cybersecurity, human oversight, transparency across multi-party action chains, and runtime behavioral drift. We propose a twelve-step compliance architecture and a regulatory trigger mapping connecting agent actions to applicable legislation. We conclude that high-risk agentic systems with untraceable behavioral drift cannot currently satisfy the AI Act's essential requirements, and that the provider's foundational compliance task is an exhaustive inventory of the agent's external actions, data flows, connected systems, and affected persons.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper provides the first systematic regulatory mapping for providers of AI agents (autonomous systems that plan, invoke tools, and execute multi-step actions) under the EU AI Act (Regulation 2024/1689) and intersecting legislation including GDPR, CRA, DSA, and NIS2. It introduces a nine-category taxonomy of agent deployment scenarios that maps concrete actions to regulatory triggers, identifies agent-specific challenges in cybersecurity, human oversight, transparency in multi-party chains, and runtime behavioral drift, and proposes a twelve-step compliance architecture. The central conclusion is that high-risk agentic systems with untraceable behavioral drift cannot satisfy the AI Act's essential requirements, with the provider's foundational task being an exhaustive inventory of external actions, data flows, connected systems, and affected persons; the analysis draws on draft harmonised standards under M/613 (Jan 2026), GPAI Code of Practice (July 2025), CRA M/606 (April 2025), and Digital Omnibus proposals (Nov 2025).

Significance. If the interpretive mappings to the cited drafts prove accurate to the final enforceable texts, the paper would offer significant practical value by translating overlapping EU obligations into a concrete taxonomy and compliance roadmap tailored to agentic systems. The explicit linkage of agent actions to regulatory triggers and the twelve-step architecture provide actionable structure for providers facing multi-party chains and drift risks, filling a gap in current guidance. The work also correctly flags that inventory of actions and data flows is a prerequisite for risk management and transparency obligations.

major comments (2)
  1. [Abstract and Conclusion] Abstract and concluding section: The claim that 'high-risk agentic systems with untraceable behavioral drift cannot currently satisfy the AI Act's essential requirements' is load-bearing for the paper's contribution, yet it is derived entirely from the specific versions of the draft harmonised standards (M/613 as of January 2026, GPAI Code July 2025, CRA M/606 April 2025, Digital Omnibus November 2025). The manuscript contains no sensitivity analysis or discussion of how alterations to final texts on human oversight, multi-party transparency, or acceptable risk-management measures would affect this conclusion, undermining the durability of the central claim.
  2. [Twelve-step compliance architecture] Section on the twelve-step compliance architecture: The architecture treats the exhaustive inventory of external actions, data flows, connected systems, and affected persons as the 'foundational compliance task,' but does not demonstrate how this inventory would be maintained dynamically in the presence of runtime behavioral drift or tool invocation chains; without addressing update mechanisms or auditability, the architecture risks being incomplete for the very drift scenarios identified as disqualifying.
minor comments (2)
  1. [Abstract] The abstract lists specific dates for the draft documents (e.g., 'as of January 2026'); these should be cross-referenced in a dedicated table or footnote in the main text to allow readers to locate the exact versions used.
  2. [Taxonomy of Agent Deployment Categories] The nine-category taxonomy is presented without concrete real-world deployment examples or case studies; adding one or two illustrative scenarios per category would improve clarity of how actions map to triggers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for their insightful review and for acknowledging the practical value of our regulatory mapping and compliance architecture for AI agent providers. We have addressed each major comment below, proposing targeted revisions to strengthen the manuscript's durability and completeness.

read point-by-point responses

  1. Referee: [Abstract and Conclusion] Abstract and concluding section: The claim that 'high-risk agentic systems with untraceable behavioral drift cannot currently satisfy the AI Act's essential requirements' is load-bearing for the paper's contribution, yet it is derived entirely from the specific versions of the draft harmonised standards (M/613 as of January 2026, GPAI Code July 2025, CRA M/606 April 2025, Digital Omnibus November 2025). The manuscript contains no sensitivity analysis or discussion of how alterations to final texts on human oversight, multi-party transparency, or acceptable risk-management measures would affect this conclusion, undermining the durability of the central claim.

    Authors: We concur that the absence of a sensitivity analysis limits the long-term applicability of our central claim. In the revised manuscript, we will insert a new paragraph in the conclusion that discusses the potential effects of changes to the final versions of the referenced standards and codes. We will emphasize that the AI Act's core obligations for risk management (Article 9) and human oversight (Article 14) require traceability and verifiability, which untraceable drift fundamentally precludes. While specific implementation details in harmonised standards may shift, the structural incompatibility is likely to persist. This addition will qualify our conclusion appropriately without altering its substance. revision: partial

  2. Referee: [Twelve-step compliance architecture] Section on the twelve-step compliance architecture: The architecture treats the exhaustive inventory of external actions, data flows, connected systems, and affected persons as the 'foundational compliance task,' but does not demonstrate how this inventory would be maintained dynamically in the presence of runtime behavioral drift or tool invocation chains; without addressing update mechanisms or auditability, the architecture risks being incomplete for the very drift scenarios identified as disqualifying.

    Authors: This is a valid critique of our architecture's presentation. To rectify this, we will revise the relevant section to explicitly outline dynamic maintenance procedures. These will include continuous logging of tool invocations, automated detection of behavioral deviations using monitoring agents, periodic re-inventory protocols triggered by detected changes, and requirements for auditable records of all updates. We will also note that these measures aim to mitigate drift but may not fully resolve untraceable cases, thereby aligning with and reinforcing our conclusion that certain high-risk systems cannot meet the essential requirements. This will render the architecture more actionable for providers. revision: yes

Circularity Check

0 steps flagged

No significant circularity; analysis draws directly from external regulatory texts.

full rationale

The paper performs a regulatory mapping and compliance analysis for AI agents by referencing external sources including the EU AI Act (Regulation 2024/1689), draft harmonised standards under M/613, the GPAI Code of Practice, CRA standards under M/606, and Digital Omnibus proposals. It introduces a nine-category taxonomy and twelve-step architecture as interpretive tools derived from these documents rather than from any internal equations, fitted parameters, or self-referential definitions. No self-citation chains, uniqueness theorems imported from prior author work, or renamings of known results appear as load-bearing steps. The central conclusion that high-risk agentic systems cannot satisfy essential requirements follows from the cited obligations on human oversight, transparency, and risk management, remaining self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper rests on the assumption that current draft standards and proposals will translate into binding requirements for AI agents without major changes, plus standard legal premises about the scope of the AI Act and related directives.

axioms (1)
  • domain assumption The EU AI Act risk-based framework and listed overlapping regulations apply directly to autonomous multi-step AI agents as described.
    Invoked throughout the abstract when mapping agent actions to regulatory triggers.

pith-pipeline@v0.9.0 · 5631 in / 1255 out tokens · 48821 ms · 2026-05-10T19:36:11.363744+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Governing What the EU AI Act Excludes: Accountability for Autonomous AI Agents in Smart City Critical Infrastructure

    cs.CY 2026-05 unverdicted novelty 5.0

    The EU AI Act narrows accountability for multi-agent AI in critical infrastructure by excluding safety components from key explanation and impact assessment rights, and the paper proposes AgentGov-SC, a three-layer ar...

  2. Dive into Claude Code: The Design Space of Today's and Future AI Agent Systems

    cs.SE 2026-04 unverdicted novelty 5.0

    Claude Code centers on a model-tool while-loop surrounded by permission systems, context compaction, extensibility hooks, subagent delegation, and session storage; the same design questions yield different answers in ...

  3. Decision Evidence Maturity Model for Agentic AI: A Property-Level Method Specification

    cs.CY 2026-04 unverdicted novelty 4.0

    DEMM defines four executable evidence-sufficiency categories plus a conflicting category for agentic AI decisions and rolls per-property verdicts into a five-level maturity rubric.

  4. Making AI Compliance Evidence Machine-Readable

    cs.CY 2026-04 unverdicted novelty 4.0

    OSCAL is extended with 16 AI-specific properties and a three-layer Compliance-as-Code architecture to generate validated assurance evidence automatically as a byproduct of training high-risk AI systems.

Reference graph

Works this paper leans on

126 extracted references · 21 canonical work pages · cited by 4 Pith papers · 4 internal anchors

  1. [1]

    Formal Opinion 512: Generative Artificial Intelligence Tools

    American Bar Association, Standing Committee on Ethics and Professional Responsibility. “Formal Opinion 512: Generative Artificial Intelligence Tools.” 29 July 2024.https://www.americanbar.org/content/dam/aba/ administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512. pdf

    2024

  2. [2]

    Agentic Artificial Intelligence from the perspective of Data Protection

    Agencia Española de Protección de Datos (AEPD). “Agentic Artificial Intelligence from the perspective of Data Protection.” 18 February 2026. https://www.aepd.es/en/guides/agentic-artificial-intelligence. pdf

    2026

  3. [3]

    Warning on highly autonomous AI agents and GDPR accountability

    Autoriteit Persoonsgegevens (Dutch Data Protection Authority). Warning on highly autonomous AI agents and GDPR accountability. February 2026. https://www.autoriteitpersoonsgegevens.nl/en/current/ ap-warns-of-major-security-risks-with-ai-agents-like-openclaw

    2026

  4. [4]

    Systemic Risks Associated with Agentic AI: A Policy Brief

    Bellogin, A., Giudici, P., Larsson, S., Pang, J., Schimpf, G., Sengupta, B., & Solmaz, G. (2025). “Systemic Risks Associated with Agentic AI: A Policy Brief.” ACM Europe TPC-Autonomous Systems Subcommit- tee. 15 October 2025. https://www.acm.org/binaries/content/assets/public-policy/europe-tpc/ systemic_risks_agentic_ai_policy-brief_final.pdf

    2025

  5. [5]

    Artificial Intelligence and Civil Liability – A European Perspective

    Bertolini, A. et al. “Artificial Intelligence and Civil Liability – A European Perspective.” European Parliament commissioned study (PE 776.426), 24 July 2025. https://www.europarl.europa.eu/RegData/etudes/ STUD/2025/776426/IUST_STU(2025)776426_EN.pdf

    2025

  6. [6]

    On the Opportunities and Risks of Foundation Models

    Bommasani, R. et al. “On the Opportunities and Risks of Foundation Models.” arXiv:2108.07258 [cs.LG], August 2021.https://arxiv.org/abs/2108.07258

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  7. [7]

    Certifying Legal AI Assistants for Unrepresented Litigants: A Global Survey of Access to Civil Justice, Unauthorized Practice of Law, and AI

    Bonardi, M. & Branting, L. “Certifying Legal AI Assistants for Unrepresented Litigants: A Global Survey of Access to Civil Justice, Unauthorized Practice of Law, and AI.” 26Science and Technology Law Review1 (2025). https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4901658

    2025

  8. [8]

    SB 53, Transparency in Frontier Artificial Intelligence Act

    California Legislature. SB 53, Transparency in Frontier Artificial Intelligence Act. Chapter 138, Statutes of

  9. [9]

    Effective 1 January 2026.https://leginfo.legislature.ca.gov/faces/billTextClient.xhtml? bill_id=202520260SB53

    2026

  10. [10]

    Civil Rights Council Secures Approval for Regulations to Protect Against Employment Discrimination Related to Artifi- cial Intelligence

    California Civil Rights Department. “Civil Rights Council Secures Approval for Regulations to Protect Against Employment Discrimination Related to Artifi- cial Intelligence.” 30 June 2025. https://calcivilrights.ca.gov/2025/06/30/ civil-rights-council-secures-approval-for-regulations-to-protect-against-employment-discrimination-related-to-artificial-intelligence/

    2025

  11. [11]

    The 2025 AI Agent Index: Documenting Technical and Safety Features of Deployed Agentic AI Systems

    Casper, S., Kolt, N. et al. “The 2025 AI Agent Index: Documenting Technical and Safety Features of Deployed Agentic AI Systems.” 2025.https://arxiv.org/abs/2602.17753

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  12. [12]

    CDT Europe’s AI Bulletin: March 2026

    Center for Democracy and Technology Europe. “CDT Europe’s AI Bulletin: March 2026.” 26 March 2026. https://cdt.org/insights/cdt-europes-ai-bulletin-march-2026/

    2026

  13. [13]

    Artificial Intelligence

    CEN-CENELEC JTC 21. “Artificial Intelligence.” https://www.cencenelec.eu/areas-of-work/ cen-cenelec-topics/artificial-intelligence/

  14. [14]

    Update on CEN and CENELEC’s Decision to Accelerate the Development of Standards for Artificial Intelligence

    CEN-CENELEC. “Update on CEN and CENELEC’s Decision to Accelerate the Development of Standards for Artificial Intelligence.” 23 October 2025. https://www.cencenelec.eu/news-events/news/2025/ brief-news/2025-10-23-ai-standardization/

    2025

  15. [15]

    Cyber Resilience Act: Standardization Request Officially Accepted by CEN, CEN- ELEC, and ETSI

    CEN-CENELEC. “Cyber Resilience Act: Standardization Request Officially Accepted by CEN, CEN- ELEC, and ETSI.” 3 April 2025. https://www.cencenelec.eu/news-events/news/2025/newsletter/ ots-62-cra/

    2025

  16. [16]

    Standardisation Request M/606 (Cyber Resilience Act)

    CEN/CENELEC/ETSI. Standardisation Request M/606 (Cyber Resilience Act). Accepted 3 April 2025. https: //www.cencenelec.eu/news-events/news/2025/newsletter/ots-62-cra/

    2025

  17. [17]

    European Commission. Standardisation Request to the European Committee for Standardisation and the Eu- ropean Committee for Electrotechnical Standardisation in support of Union policy on artificial intelligence. C(2023) 3215 final (M/593, as amended by M/613). https://ec.europa.eu/growth/tools-databases/ enorm/mandate/593_en

    2023

  18. [18]

    prEN 18228: Artificial Intelligence – Risk Management

    CEN/CENELEC JTC 21. prEN 18228: Artificial Intelligence – Risk Management. Working draft, Jan- uary 2026. https://standards.cencenelec.eu/ords/f?p=205:110:::::FSP_PROJECT,FSP_LANG_ID: 79438,25&cs=126AEABA70EBCF2433A6A5472A8FD6F84 44 AI AGENTSUNDEREU LAW- WORKINGPAPER, APRIL7, 2026

    2026

  19. [19]

    prEN 18229-1: Artificial Intelligence – AI trustworthiness framework – Part 1: Logging, transparency and human oversight

    CEN/CENELEC JTC 21. prEN 18229-1: Artificial Intelligence – AI trustworthiness framework – Part 1: Logging, transparency and human oversight. Working draft, January 2026.https://standards.cencenelec.eu/ords/ f?p=205:110:::::FSP_PROJECT,FSP_LANG_ID:76986,25&cs=17E4F9ABCBAC14D2D4C9D36D274FAA1FB

    2026

  20. [20]

    prEN 18229-2: Artificial Intelligence – Trustworthiness Framework – Part 2: Accu- racy and Robustness

    CEN/CENELEC JTC 21. prEN 18229-2: Artificial Intelligence – Trustworthiness Framework – Part 2: Accu- racy and Robustness. Working draft, January 2026. https://standards.cencenelec.eu/ords/f?p=205: 110:::::FSP_PROJECT,FSP_LANG_ID:82493,25&cs=1400FEB0B6AA9D5AB34BF0233CC4E75B7

    2026

  21. [21]

    prEN 18281: Artificial Intelligence — Evaluation methods for accurate computer vi- sion systems

    CEN/CENELEC JTC 21. prEN 18281: Artificial Intelligence — Evaluation methods for accurate computer vi- sion systems. Working draft, January 2026. https://standards.cencenelec.eu/ords/f?p=205:110::::: FSP_PROJECT,FSP_LANG_ID:79657,25&cs=14D236550058D93DCA7973A19C8D13B24

    2026

  22. [22]

    prEN 18282: Artificial Intelligence – Cybersecurity specifications for AI Sys- tems

    CEN/CENELEC JTC 21. prEN 18282: Artificial Intelligence – Cybersecurity specifications for AI Sys- tems. Working draft, January 2026. https://standards.cencenelec.eu/ords/f?p=205:110:::::FSP_ PROJECT,FSP_LANG_ID:79708,25&cs=1F22C53E33572EA17236E5EF8F9DE9DCD

    2026

  23. [23]

    prEN 18283: Artificial Intelligence — Concepts, measures and requirements for managing bias in AI systems

    CEN/CENELEC JTC 21. prEN 18283: Artificial Intelligence — Concepts, measures and requirements for managing bias in AI systems. Working draft, January 2026. https://standards.cencenelec.eu/ords/f? p=205:110:::::FSP_PROJECT,FSP_LANG_ID:80353,25&cs=1A0A0C0DBC7B012D69EA9AAAF8D5DFDA3

    2026

  24. [24]

    CEN/CENELEC JTC 21. prEN 18284: Artificial Intelligence — Quality and governance of datasets in AI, January 2026 https://standards.cencenelec.eu/ords/f?p=205:110:::::FSP_PROJECT,FSP_LANG_ ID:80364,25&cs=16B97AF755CF36534FF119CB9782A0D1C

    2026

  25. [25]

    prEN 18286: Artificial intelligence - Quality management system for EU AI Act regulatory purposes

    CEN/CENELEC JTC 21. prEN 18286: Artificial intelligence - Quality management system for EU AI Act regulatory purposes. Working draft, January 2026. https://standards.cencenelec.eu/ords/f?p=205: 110:::::FSP_PROJECT,FSP_LANG_ID:80556,25&cs=175392A97352F2C5B1211EEC4FA215C15

    2026

  26. [26]

    SB 24-205, Concerning Consumer Protections for Artificial Intelligence

    Colorado General Assembly. SB 24-205, Concerning Consumer Protections for Artificial Intelligence. Signed 17 May 2024. Effective date delayed to 30 June 2026 by SB 25B-004. https://leg.colorado.gov/bill_ files/47770/download

    2024

  27. [27]

    AI Visions in 2026: A Transatlantic Strategic Divide

    Control Risks. “AI Visions in 2026: A Transatlantic Strategic Divide.” Early 2026.https://www.controlrisks. com/our-thinking/insights/ai-visions-in-2026-a-transatlantic-strategic-divide

    2026

  28. [28]

    HUDERIA: Human Rights, Democracy and Rule of Law Impact Assessment for AI Systems

    Council of Europe. “HUDERIA: Human Rights, Democracy and Rule of Law Impact Assessment for AI Systems.” Developed under the Framework Convention on AI (CETS No. 225), 2025.https://www.coe.int/en/web/ artificial-intelligence/huderia

    2025

  29. [29]

    General approach on the proposal for a Regulation amending Regula- tion (EU) 2024/1689 (Digital Omnibus on AI)

    Council of the European Union. General approach on the proposal for a Regulation amending Regula- tion (EU) 2024/1689 (Digital Omnibus on AI). 13 March 2026. https://www.consilium.europa.eu/en/ policies/digital-agenda/eu-ai-act/

    2024

  30. [30]

    Systemic Risks of Interacting AI

    Darius, P., Hoppe, T., and Aleksandrov, A. “Systemic Risks of Interacting AI.” arXiv:2512.17793 [cs.AI], December 2025.https://arxiv.org/abs/2512.17793

    work page arXiv 2025

  31. [31]

    Algorithmic discrimination under the AI Act and the GDPR

    De Luca, S. “Algorithmic discrimination under the AI Act and the GDPR.” European Parliamentary Research Ser- vice, 26 February 2025.https://www.europarl.europa.eu/thinktank/en/document/EPRS_ATA(2025) 769509

    2025

  32. [32]

    AI Agents Under Threat: A Survey of Key Security Challenges and Future Pathways

    Deng, Z., Guo, Y ., Han, C., Ma, W., Xiong, J., Wen, S., & Xiang, Y . “AI Agents Under Threat: A Survey of Key Security Challenges and Future Pathways.”ACM Computing Surveys, 57(7), 1-36. 2025. https://doi.org/10. 1145/3716628

    2025

  33. [33]

    Regulatory Innovation at the Cross- roads: Five Years of Data on Entity-Regulation Reform in Arizona and Utah

    Engstrom, D.F., Knowlton, L. & Ricca, D. “Regulatory Innovation at the Cross- roads: Five Years of Data on Entity-Regulation Reform in Arizona and Utah.” Stanford Law School, 2 June 2025. https://law.stanford.edu/2025/06/02/ regulatory-innovation-at-the-crossroads-five-years-of-data-on-entity-regulation-reform-in-arizona-and-utah/

    2025

  34. [34]

    Security and Privacy Considerations in Autonomous Agents

    ENISA. “Security and Privacy Considerations in Autonomous Agents.” https://www.enisa.europa.eu/ publications/considerations-in-autonomous-agents

  35. [35]

    Cyber Resilience Act Requirements Standards Mapping

    ENISA and JRC. “Cyber Resilience Act Requirements Standards Mapping.” 2024/2025.https://www.enisa. europa.eu/publications/cyber-resilience-act-requirements-standards-mapping

    2024

  36. [36]

    European Telecommunications Standards Institute, December 2025.https: //www.etsi.org/deliver/etsi_en/304200_304299/304223/02.01.01_60/en_304223v020101p.pdf

    ETSI TC CYBER SAI.ETSI EN 304 223 V2.1.1: Securing Artificial Intelligence (SAI); Baseline Cyber Security Re- quirements for AI Models and Systems. European Telecommunications Standards Institute, December 2025.https: //www.etsi.org/deliver/etsi_en/304200_304299/304223/02.01.01_60/en_304223v020101p.pdf

    2025

  37. [37]

    Working draft under CRA Standardisation Request M/606, available under Open Consultation

    ETSI TC CYBER EUSR.ETSI EN 304 617 (draft): Cybersecurity requirements for browsers with digital elements. Working draft under CRA Standardisation Request M/606, available under Open Consultation. https: //docbox.etsi.org/CYBER/EUSR/Open 45 AI AGENTSUNDEREU LAW- WORKINGPAPER, APRIL7, 2026

    2026

  38. [38]

    Frequently Asked Questions

    European AI Office. “Frequently Asked Questions.” AI Act Service Desk, European Commission. https: //ai-act-service-desk.ec.europa.eu/en/faq(accessed March 2026)

    2026

  39. [39]

    Code of Practice on Marking and Labelling of AI-generated Con- tent — Second Draft

    European Commission / AI Office. “Code of Practice on Marking and Labelling of AI-generated Con- tent — Second Draft.” 5 March 2026. https://digital-strategy.ec.europa.eu/en/library/ commission-publishes-second-draft-code-practice-marking-and-labelling-ai-generated-content

    2026

  40. [40]

    Blue Guide on the implementation of EU product rules 2022

    European Commission. Blue Guide on the implementation of EU product rules 2022. OJ C 247/01, 29.6.2022. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=OJ:C:2022:247:TOC

    2022

  41. [41]

    Commission calls on 10 Member States to comply with the Data Governance Act

    European Commission. “Commission calls on 10 Member States to comply with the Data Governance Act.” 16 December 2024. https://digital-strategy.ec.europa.eu/en/news/ commission-calls-10-member-states-comply-data-governance-act

    2024

  42. [42]

    COM(2020) 64 final

    European Commission. COM(2020) 64 final. Report on the safety and liability implications of Artificial Intelli- gence, the Internet of Things and robotics. 19 February 2020.https://eur-lex.europa.eu/legal-content/ EN/TXT/PDF/?uri=CELEX:52020DC0064

    2020

  43. [43]

    Agentic AI: Leveraging European AI Talent and Regulatory Assets to Scale Adoption

    European Commission. “Agentic AI: Leveraging European AI Talent and Regulatory Assets to Scale Adoption.” StepUp StartUps Initiative, 23 January 2026. https://digital-strategy.ec.europa.eu/en/library/ agentic-ai-leveraging-european-ai-talent-and-regulatory-assets-scale-adoption

    2026

  44. [44]

    Guidelines on the scope of obligations for providers of general-purpose AI models under the AI Act

    European Commission. “Guidelines on the scope of obligations for providers of general-purpose AI models under the AI Act.” 2025. https://digital-strategy.ec.europa.eu/en/library/ guidelines-scope-obligations-providers-general-purpose-ai-models-under-ai-act

    2025

  45. [45]

    COM(2025) 836 final

    European Commission. COM(2025) 836 final. Proposal for a Regulation amending Regulation (EU) 2024/1689 (Digital Omnibus on AI). 19 November 2025. https://eur-lex.europa.eu/legal-content/EN/TXT/ ?uri=celex:52025PC0836

    2025

  46. [46]

    COM(2025) 837 final

    European Commission. COM(2025) 837 final. Proposal for a Regulation amending Regulation (EU) 2023/2854 and other digital legislation (Digital Legislation Omnibus). 19 November 2025.https://eur-lex.europa.eu/ legal-content/EN/TXT/?uri=celex:52025PC0837

    2025

  47. [47]

    Harmonised Standards for the European AI Act

    European Commission JRC / AI Watch. “Harmonised Standards for the European AI Act.” Science for Policy Brief, 25 October 2024. https://ai-watch.ec.europa.eu/news/ harmonised-standards-european-ai-act-2024-10-25_en

    2024

  48. [48]

    Agentic AI

    European Data Protection Supervisor. “Agentic AI.” TechSonar 2025–2026, 24 November 2025.https://www. edps.europa.eu/data-protection/technology-monitoring/techsonar/agentic-ai_en

    2025

  49. [49]

    Report A10-0073/2026 on the proposal for a Regulation amending Regulation (EU) 2024/1689 (Digital Omnibus on AI)

    European Parliament, Committees IMCO and LIBE. Report A10-0073/2026 on the proposal for a Regulation amending Regulation (EU) 2024/1689 (Digital Omnibus on AI). Adopted in committee 18 March 2026. https: //www.europarl.europa.eu/doceo/document/A-10-2026-0073_EN.html

    2026

  50. [50]

    Plenary vote validating negotiating mandate on Digital Omnibus on AI (A10-0073/2026)

    European Parliament. Plenary vote validating negotiating mandate on Digital Omnibus on AI (A10-0073/2026). 26 March 2026.https://www.europarl.europa.eu/news/en/press-room/20260323IPR38829

    work page arXiv 2026

  51. [51]

    Digital Omnibus on AI

    European Parliamentary Research Service. “Digital Omnibus on AI.” Briefing, 2026.https://www.europarl. europa.eu/thinktank/en/document/EPRS_BRI(2026)782651

    2026

  52. [52]

    Published 10 July 2025

    EU Code of Practice for General-Purpose AI Models. Published 10 July 2025. https://digital-strategy. ec.europa.eu/en/policies/contents-code-gpai

    2025

  53. [53]

    Declaratory Ruling,Implications of Artificial Intelligence Technologies on Protecting Consumers from Unwanted Robocalls and Robotexts, CG Docket No

    Federal Communications Commission. Declaratory Ruling,Implications of Artificial Intelligence Technologies on Protecting Consumers from Unwanted Robocalls and Robotexts, CG Docket No. 23-362, FCC 24-17, 39 FCC Rcd 1783 (2024). Adopted 2 February 2024. https://docs.fcc.gov/public/attachments/FCC-24-17A1.pdf

    2024

  54. [54]

    FTC Launches Inquiry into AI Chatbots Acting as Companions

    Federal Trade Commission. “FTC Launches Inquiry into AI Chatbots Acting as Companions.” Press Release, 11 September 2025. https://www.ftc.gov/news-events/news/press-releases/2025/09/ ftc-launches-inquiry-ai-chatbots-acting-companions

    2025

  55. [55]

    Children’s Online Privacy Protection Rule: Final Rule

    Federal Trade Commission. Children’s Online Privacy Protection Rule: Final Rule. 90 Fed. Reg. 16918 (22 April 2025) (codified at 16 C.F.R. Part 312). Compliance deadline: 22 April 2026. https://unblock. federalregister.gov/

    2025

  56. [56]

    232-3052

    Federal Trade Commission.In the Matter of Rytr LLC, File No. 232-3052. Final Order set aside 22 December 2025 (2–0 vote, Chairman Ferguson and Commissioner Meador).https://www.ftc.gov/legal-library/browse/ cases-proceedings/232-3052-rytr-llc-matter 46 AI AGENTSUNDEREU LAW- WORKINGPAPER, APRIL7, 2026

    2025

  57. [57]

    Policy Statement on AI and Section 5 of the FTC Act

    Federal Trade Commission. Policy Statement on AI and Section 5 of the FTC Act. 11 March 2026. https: //www.jdsupra.com/legalnews/emerging-federal-ai-policy-what-to-know-8882048/#:~: text=January%2010%2C%202026:%20DOJ%20AI,the%20date%20of%20this%20alert

    2026

  58. [58]

    Human Oversight under Article 14 of the EU AI Act

    Fink, M. “Human Oversight under Article 14 of the EU AI Act.” 2025.https://www.aigl.blog/content/ files/2025/04/Human-Oversight-under-Article-14-of-the-EU-AI-Act.pdf

    2025

  59. [59]

    National supervision of EU Artificial Intelligence Act to begin – laws on powers of authorities to take effect at start of the year

    Finnish Ministry of Economic Affairs and Employment. “National supervision of EU Artificial Intelligence Act to begin – laws on powers of authorities to take effect at start of the year.” 7 January 2026. https://tem.fi/en/-/1410877/ national-supervision-of-eu-artificial-intelligence-act-to-begin-laws-on-powers-of-authorities-to-take-effect-at-start-of-the-year

    work page arXiv 2026

  60. [60]

    Conformity Assessments under the EU AI Act: A Step-by-Step Guide

    Serban, A., Rovilos, V ., Demetzou, K. “Conformity Assessments under the EU AI Act: A Step-by-Step Guide.” Future of Privacy Forum / OneTrust, updated April 2025. https://fpf.org/wp-content/uploads/2025/ 04/OT-comformity-assessment-under-the-eu-ai-act-WP-1.pdf

    2025

  61. [61]

    Understanding the New Wave of Chatbot Leg- islation: California SB 243 and Beyond

    Future of Privacy Forum. “Understanding the New Wave of Chatbot Leg- islation: California SB 243 and Beyond.” 2025. https://fpf.org/blog/ understanding-the-new-wave-of-chatbot-legislation-california-sb-243-and-beyond/

    2025

  62. [62]

    Regulating AI Agents

    Gardhouse, K., Oueslati, A., & Kolt, N. “Regulating AI Agents.” Working paper, 24 March 2026. https: //arxiv.org/abs/2603.23471

    work page arXiv 2026

  63. [63]

    Interplay between the AI Act and the EU digital legislative framework

    Graux, H., Garstka, K., Murali, N., Cave, J., Botterman, M. “Interplay between the AI Act and the EU digital legislative framework.” European Parliament, ITRE Committee, Policy Department for Transformation, Innovation and Health, 29–30 October 2025. https://www.europarl.europa.eu/thinktank/en/document/ECTI_ STU(2025)778575

    2025

  64. [64]

    Multi-agent risks from advanced ai.arXiv preprint arXiv:2502.14143, 2025

    Hammond, L., Chan, A., Clifton, J. et al. “Multi-Agent Risks from Advanced AI.” Cooperative AI Foundation Technical Report, 19 February 2025.https://arxiv.org/abs/2502.14143

    work page arXiv 2025

  65. [65]

    EU Cyber Resilience Act: Key 2026 Milestones toward CRA Compliance

    Hanssen, H. and Brockmeyer, J. “EU Cyber Resilience Act: Key 2026 Milestones toward CRA Compliance.” Hogan Lovells, 20 January 2026. https://www.hoganlovells.com/en/publications/ eu-cyber-resilience-act-getting-ready-for-cra-compliance-in-2026

    2026

  66. [66]

    EU AI Act Regulatory Directory

    IAPP. “EU AI Act Regulatory Directory.” Updated January 2026.https://iapp.org/resources/article/ eu-ai-act-regulatory-directory

    2026

  67. [67]

    European Commission misses deadline for AI Act guidance on high-risk systems

    IAPP. “European Commission misses deadline for AI Act guidance on high-risk systems.” 3 February 2026. https://iapp.org/news/a/ european-commission-misses-deadline-for-ai-act-guidance-on-high-risk-systems

    2026

  68. [68]

    Model AI Governance Framework for Agentic AI

    Infocomm Media Development Authority (IMDA), Singapore. “Model AI Governance Framework for Agentic AI.” 22 January 2026. https://www.imda.gov.sg/-/media/imda/files/about/ emerging-tech-and-research/artificial-intelligence/mgf-for-agentic-ai.pdf

    2026

  69. [69]

    Colorado Officials Push to Repeal and Replace the Col- orado AI Act

    Inside Global Tech. “Colorado Officials Push to Repeal and Replace the Col- orado AI Act.” 27 March 2026. https://www.insideglobaltech.com/2026/03/27/ colorado-officials-push-to-repeal-and-replace-the-colorado-ai-act/

    2026

  70. [70]

    2025.https://www.iso.org/standard/89455.html

    ISO/IEC 4213 ed.2: Artificial intelligence — Performance measurement for AI classification, regression, clustering and recommendation tasks. 2025.https://www.iso.org/standard/89455.html

    2025

  71. [71]

    Published 2025.https://www.iso.org/standard/84111.html

    ISO/IEC 12792:2025: Information technology — Artificial intelligence — Transparency taxonomy of AI systems. Published 2025.https://www.iso.org/standard/84111.html

    2025

  72. [72]

    FDIS registered 12 March 2026

    ISO/IEC FDIS 27090: Cybersecurity — Artificial Intelligence — Guidance for addressing security threats and compromises to artificial intelligence systems. FDIS registered 12 March 2026. https://www.iso.org/ standard/56581.html

    2026

  73. [73]

    International Organization for Standardization, 2022

    ISO.ISO 31073:2022: Risk management — Vocabulary. International Organization for Standardization, 2022. https://www.iso.org/standard/79637.html

    2022

  74. [74]

    Published May 2025.https://www.iso.org/standard/42005

    ISO/IEC 42005:2025, Information technology — Artificial intelligence — AI system impact assessment. Published May 2025.https://www.iso.org/standard/42005

    2025

  75. [75]

    DIS ballot initiated 24 November 2025.https://www.iso.org/standard/86902.html

    ISO/IEC DIS 42105: Information technology — Artificial intelligence — Guidance for human oversight of AI systems. DIS ballot initiated 24 November 2025.https://www.iso.org/standard/86902.html

    2025

  76. [76]

    prEN ISO/IEC 23282: Artificial Intelligence - Evaluation methods for accurate natural language processing systems Draft

    ISO/IEC. prEN ISO/IEC 23282: Artificial Intelligence - Evaluation methods for accurate natural language processing systems Draft. https://standards.cencenelec.eu/ords/f?p=205:110:::::FSP_PROJECT, FSP_LANG_ID:77582,25&cs=12CADCFE745035835A4EE315C44A5045C 47 AI AGENTSUNDEREU LAW- WORKINGPAPER, APRIL7, 2026

    2026

  77. [77]

    prEN ISO/IEC 24970: Artificial intelligence - AI system logging (ISO/IEC DIS 24970:2025)

    ISO/IEC. prEN ISO/IEC 24970: Artificial intelligence - AI system logging (ISO/IEC DIS 24970:2025). Draft https://standards.cencenelec.eu/ords/f?p=205:110:::::FSP_PROJECT,FSP_LANG_ID: 78565,25&cs=11492185C74B08825BC0990EEEF862587

    2025

  78. [78]

    A Comparative Analysis of the EU AI Act and the Colorado AI Act: Regulatory Approaches to Artificial Intelligence Governance

    Jariwala, M. “A Comparative Analysis of the EU AI Act and the Colorado AI Act: Regulatory Approaches to Artificial Intelligence Governance.”Int’l J. Computer Applications, 186(38): 23–29, September 2024. https: //doi.org/10.5120/ijca2024923954

    work page doi:10.5120/ijca2024923954 2024

  79. [79]

    Preprint at https://arxiv.org/abs/ 2601.11893

    Ji, Y . et al. “Taming Various Privilege Escalation in LLM-Based Agent Systems: A Mandatory Access Control Framework.” arXiv:2601.11893 [cs.CR], January 2026.https://arxiv.org/abs/2601.11893

    work page arXiv 2026

  80. [80]

    Agentic Tool Sovereignty

    Jones, L. “Agentic Tool Sovereignty.” European Law Blog, 2025.https://www.europeanlawblog.eu/pub/ dq249o3c

    2025

Showing first 80 references.