Privacy & Cybersecurity #73
EU DMA Report | UK AI Priorities | Spain AI Governance Law | AI Conversation-Tracking Risks | Stockholm Declaration | Canada on AI | Vermont Neuro Rights | Pope Leo on AI | NIST Report | OECD on CS
May 31, 2026
🇪🇺 EU Commission Reports First Full Enforcement Year Under the Digital Markets Act
On 21 May 2026, the European Commission published its third annual report on implementation of the Digital Markets Act (DMA), covering enforcement activity in 2025. The report confirms that, by the end of 2025, seven gatekeepers remained subject to the DMA: Alphabet, Amazon, Apple, Booking, ByteDance, Meta and Microsoft, covering 23 core platform services.
The report marks 2025 as the first year in which DMA enforcement produced visible regulatory outcomes. The Commission emphasized regulatory dialogue, compliance monitoring, market investigations, and formal non-compliance decisions. The year also brought the first removal of a core platform service from a designation decision: Meta Marketplace was removed after the Commission concluded that it no longer met the DMA’s quantitative business-user threshold.
A central enforcement theme was the treatment of user data. The Commission adopted a non-compliance decision against Meta under Article 5(2) DMA in relation to its “consent or pay” model for Facebook and Instagram. The Commission found that Meta’s model did not give users the required option to choose a less personalized but equivalent service where their personal data would not be combined across services for personalized advertising. Meta was fined EUR 200 million.
The Commission also fined Apple EUR 500 million for restrictions affecting app developers’ ability to steer users to offers outside Apple’s App Store. The Commission found that Apple’s rules prevented developers from fully informing users about alternative and potentially cheaper offers outside the App Store, and ordered Apple to remove the relevant technical and commercial restrictions.
In March 2025, the Commission adopted two legally binding specification decisions concerning Apple’s obligations under Article 6(7) DMA. These decisions set conditions for interoperability between iOS and connected devices such as smartwatches, headphones and other wearables, and also set rules for Apple’s process for handling interoperability requests. The measures cover access to connectivity features, faster data transfers, easier device pairing, improved notifications, technical documentation, and more predictable request timelines.
The report also records progress on user choice in mobile ecosystems. Apple improved browser choice screens, made it easier to change default settings on iOS, and allowed certain pre-installed Apple applications, including Safari and the App Store, to be uninstalled. Microsoft changed Windows-related settings to make it easier for users to set and use a default browser. Alphabet continued to roll out browser and search-engine choice screens, and dropped the requirement that users create or use a Gmail account to access Google services such as Android, YouTube or Google Play.
On 18 November 2025, the Commission opened two market investigations into whether Amazon Web Services and Microsoft Azure should be designated as gatekeeper services under the DMA, despite not meeting the standard quantitative thresholds. The Commission also opened a broader cloud-sector investigation into whether the DMA’s existing obligations are effective in addressing unfair or anti-contestable practices in cloud computing. The investigation covers issues such as interoperability obstacles, restricted access to data, tying and bundling, and potentially imbalanced contract terms. The report says this broader investigation should conclude within 18 months and may lead to proposed updates to DMA obligations for cloud services.
The report notes that cloud computing is essential for businesses operating in the EU and for the development of AI technologies. This indicates that the DMA may increasingly be used to examine infrastructure-layer dependencies, not only consumer-facing platforms such as app stores, search engines, browsers and social networks.
The report also highlights merger transparency under Article 14 DMA. In 2025, gatekeepers submitted 36 notifications of intended concentrations, bringing the total since 2023 to 55. The Commission observed that many 2025 notifications concerned AI-related transactions, including “acqui-hiring” arrangements involving the acquisition of talent. This gives regulators better visibility into strategic acquisitions by major digital platforms, even where a transaction may not automatically trigger traditional merger-control thresholds.
Gatekeepers submitted updated independently audited descriptions of consumer profiling techniques under Article 15 DMA. These reports are shared with the European Data Protection Board and are intended to improve transparency around profiling practices across designated core platform services. The Commission and the EDPB also worked on joint guidelines concerning the interplay between the DMA and the GDPR, with adoption expected in 2026.
🇬🇧UK ICO Sets 2026/27 AI Priorities: Code of Practice, Agentic AI Guidance, and Transparency Tools
On 27 May 2026, the UK Information Commissioner’s Office (ICO) wrote to the Secretaries of State for Science, Innovation and Technology and for Business and Trade outlining its planned work on AI-driven innovation. The letter responds to a 28 January 2026 government request asking the ICO to explain how it will support safe AI-powered innovation and report on that work.
The ICO frames its AI work around three linked aims: supporting business and public-sector use of AI, clarifying the standards and safeguards expected under data protection law, and ensuring that individuals retain transparency and control where their personal data is processed by AI systems or used in significant automated decisions.
ICO is preparing an updated AI workplan for 2026/27. That workplan will have two overarching objectives. First, the ICO wants the UK public to understand how AI systems process their personal data and to have appropriate agency, choice, and control over that use. Second, the ICO wants organizations to have clearer guidance on what data protection law requires when they deploy AI systems, including AI agents and agentic systems.
The planned actions include several items that will matter directly to compliance teams.
The ICO will develop an AI and automated decision-making statutory code of practice. It should help translate UK GDPR and Data Protection Act requirements into operational expectations around transparency, lawful basis, fairness, accountability, individual rights, and safeguards for significant automated decisions.
The ICO also intends to publish guidance on how agentic systems can comply with UK GDPR. This is significant because agentic AI raises difficult governance issues: systems may act with a degree of autonomy, use personal data across multiple steps, interact with external tools, and generate outputs that are not always predictable at the design stage.
For procurement, the ICO plans to publish a transparency resource for organizations, particularly SMEs and public bodies, to support data protection due diligence when buying “off the shelf” cloud-based AI tools and services. This may become especially useful for vendor assessment, DPIAs, records of processing, contractual review, and internal approval of AI tools.
The ICO also plans public-facing guidance: a “green cross code-style” guide to help individuals make informed decisions about how online AI tools and services use their personal data. In parallel, the ICO says it will engage with major technology companies on the growing personalization of consumer-facing AI services, with a focus on transparency and privacy-focused product design.
The ICO will streamline and rebrand its Innovation and Sandbox services to make them easier for organizations developing and deploying AI to access.
🇫🇷 CNIL Highlights Breach Response Duties When a Subcontractor Is Attacked
The French data protection authority, CNIL, has published an educational scenario explaining how organizations should respond when a subcontractor providing a cloud service suffers a cyberattack that affects both its own data and client data.
A cybercriminal targets a subcontractor that provides a cloud solution to business clients. Using social engineering, the attacker deceives an employee, gains access to the subcontractor’s information system, and copies a large volume of data belonging not only to the subcontractor but also to its client companies. The incident is detected through monitoring tools, after which the security lead confirms the data leak and activates the internal incident response procedure. The system is put into protective mode, causing a service blackout for the subcontractor and its clients.
The central point is the dual role of the subcontractor. It is a controller for its own data, including employee and corporate data. At the same time, it acts as a processor for the data it handles on behalf of its clients. These roles create different breach response obligations.
For its own data, the subcontractor must assess what data was affected, whether personal data was involved, and the level of risk for the individuals concerned. If the breach is notifiable, the organization must notify the CNIL within 72 hours. This reflects the GDPR rule that controllers must notify the competent supervisory authority without undue delay and, where feasible, within 72 hours after becoming aware of a personal data breach, unless the breach is unlikely to result in a risk to individuals’ rights and freedoms.
For client data, the subcontractor’s first duty is to inform the affected client companies as quickly as possible so that they can meet their own legal obligations. The processor does not replace the controller’s responsibility. Each client company must assess the breach in relation to the data it controls, determine whether supervisory authority notification is required, and decide whether affected individuals must be informed.
The CNIL scenario also shows what good operational support from a processor can look like. The subcontractor sends each affected client an information message, provides a guide to assist with breach notification, opens a dedicated telephone line for questions, and offers to submit notifications on behalf of clients where the client gives formal authorization.
Where the data theft creates a high risk to individuals’ rights and freedoms, each affected client company must inform the individuals concerned clearly and promptly. The notice should explain what happened, how the company responded, what data was affected, and what the possible consequences are.
🇪🇸 Spain Approves Draft Organic Law on AI Governance and Human Oversight
On 26 May 2026, Spain’s Council of Ministers approved the draft Organic Law on the proper use and governance of artificial intelligence, sending it to the Congress of Deputies. The bill is intended to adapt Spanish law to the EU AI Act, which has been in force since August 2024, and to establish Spain’s national supervisory and sanctions framework for AI systems.
The draft law designates the Spanish AI Supervisory Agency, AESIA, as a central authority for supervision, with a single contact point for supervisory matters. For AI systems already covered by sectoral product legislation, such as machinery, toys, vehicles, and medical devices, the existing notifying and market surveillance authorities will continue to supervise compliance. For other systems, including those used in employment, biometrics, and education, supervisory competence will be allocated mainly among AESIA, the Spanish Data Protection Authority, AEPD, and the General Council of the Judiciary, depending on the relevant field.
The bill also introduces a Spanish public-sector AI governance layer that goes beyond mere implementation of the EU AI Act. State public-sector bodies will be required to maintain an inventory of AI systems used in administrative procedures, not only high-risk systems. The law also creates the role of an AI delegate, responsible for coordinating legal compliance and advising on AI projects and public procurement. Both the inventory and AI delegate model will be developed further by Royal Decree.
The sanctions regime follows the structure of the EU AI Act. Infringements will be classified as very serious, serious, or minor. For the most serious cases, fines may reach EUR 35 million or 7% of annual turnover. For minor infringements, penalties may reach EUR 500,000 or 0.5% of annual turnover. The draft law also provides flexibility for regulators, including consideration of intent, recurrence, corrective measures, prompt payment, and the size of the company, with specific attention to SMEs and startups.
The draft law formalizes the national sandbox structure, to be operated by AESIA, and allows additional sector-specific sandboxes where they are created by competent market surveillance or notifying authorities. Relevant public policy bodies and fundamental rights authorities must participate where their areas are affected.
🇪🇸 AEPD Refers AI Conversation-Tracking Risks to European Data Protection Authorities
Spain’s data protection authority, the Agencia Española de Protección de Datos (AEPD), has asked European data protection authorities to study whether some AI systems allow third parties to access users’ conversations through tracking technologies. The issue will be discussed with other European authorities at meetings scheduled for 8 and 9 June 2026.
The AEPD’s action follows a study by IMDEA Networks, published as “LeakyLM - Your AI Assistant Is Leaking Your Conversations.” The study examined generative AI services, including Perplexity, Anthropic’s Claude, xAI’s Grok, and OpenAI’s ChatGPT. It reports that some services embedded third-party analytics or advertising trackers that transmitted conversation URLs, page titles, user identifiers, email hashes, or other metadata to third parties such as Google, Meta, TikTok, Datadog, Intercom, and others.
AI conversations often include health concerns, employment issues, legal questions, commercial information, source code, or other sensitive material. The IMDEA study states that conversation URLs and titles may reveal the subject matter of a user’s interaction with an AI system. In some cases, where a conversation link functions as a permalink with weak access controls, a third party receiving the URL may have the technical ability to access the conversation or infer its content.
The report identifies the issues: exposure of permanent links from conversations to third-party trackers; the possibility of linking AI interactions to user identities through cookies, email hashes, or other identifiers; and privacy controls or disclosures that may not clearly reflect actual data flows. The AEPD’s press note frames the matter as one requiring coordinated European review, taking into account the GDPR one-stop-shop mechanism.
If conversation URLs, titles, or identifiers are disclosed to advertising or analytics providers, AI service operators will need to justify the relevant processing purpose, legal basis, transparency disclosures, and data minimization controls. Where non-essential trackers are used, the ePrivacy consent framework may also be relevant. The risk is higher where users are told that conversations are private or restricted, while associated metadata or links are still transmitted to third parties through web analytics, support widgets, advertising pixels, or server-side tracking.
Many organizations now allow employees to use AI assistants for drafting, research, coding, customer support, and internal analysis. If employees enter confidential business information, personal data, client material, or regulated data into AI tools, the risk may extend beyond the AI provider’s own model training or retention practices. Tracking infrastructure, embedded scripts, support tools, and server-side event routing can become part of the relevant data flow and should be assessed in vendor due diligence.
Nordic DPAs Sign Stockholm Declaration
On 21–22 May 2026, the data protection authorities of Denmark, the Faroe Islands, Finland, Iceland, Norway, Åland, and Sweden met in Stockholm for their annual Nordic Meeting and signed the Stockholm Declaration. The Nordic Meetings have been held since 1988 and serve as a forum for cooperation and exchange of best practices among the Nordic supervisory authorities.
The declaration signals about regulatory priorities in the Nordic region: closer cooperation among DPAs, a more pragmatic approach to enforcement, and greater attention to overlapping digital regulation, including the GDPR, the AI Act, and the Data Act.
The declaration identifies several areas of cooperation.
The DPAs intend to work together on the implementation of the EU procedural rules regulation for GDPR enforcement. A more coordinated Nordic approach may help make enforcement more consistent.
The authorities highlight artificial intelligence as a core supervisory topic. They plan to share experience and best practices on internal use of AI by DPAs and external supervisory work relating to AI.
The declaration recognizes that complaint handling is becoming more complex, including because of AI-related issues. The DPAs state that they will continue to exchange best practices to handle complaints efficiently.
The declaration refers to public awareness activities and continued cooperation with the European Data Protection Board.
The DPAs emphasize the need for an “even more risk-based approach” across Europe in applying EU data protection rules, while ensuring that those rules remain relevant and credible instruments for protecting fundamental rights. The Declaration suggests that Nordic regulators will continue to support strong privacy protection, but with attention to proportionality, prioritization, and practical implementation.
🇨🇦 Canada Issues Guidance on Agentic AI Use in Federal Government
On 22 May 2026, the Government of Canada published its Guide on the Use of Agentic Artificial Intelligence, aimed at federal departments and agencies considering the use, piloting, procurement, or deployment of AI systems that can take actions.
The guidance identifies risks such as unintended system interactions, misaligned task execution, reduced auditability, unauthorized access or disclosure of sensitive information, and overreliance by employees on AI-generated recommendations.
The Government of Canada advises that agentic AI should be considered only where the intended outcomes are clearly defined, decision boundaries are explicit, accountability is assigned, and risks can be tested, monitored, and managed across the system lifecycle. Departments must also assess whether the Directive on Automated Decision-Making applies, and must consider other legal, privacy, security, human rights, language rights, and IM/IT obligations.
The guide introduces two agentic-AI-specific principles: bounded autonomy and recoverability. Bounded autonomy means that AI agents should operate within tight and explicit limits on data, tools, permissions, and scope. The guide recommends clear activity labels such as “draft only” or “read only,” documented ownership, and technical constraints that prevent unsafe actions before they occur. Recoverability means that AI agents should be capable of being paused, stopped, corrected, and returned to a safe state. The guide also recommends full logging of agent actions in a system that the agent cannot alter.
The guidance follows a lifecycle structure: before use, during use, and after deployment. Before use, departments should start with a narrow and well-defined use case, map the process, consult legal, privacy, security, IM/IT, and program experts, plan for explainability, and assess the application of the Directive on Automated Decision-Making. During use, departments should maintain named human accountability, escalation paths, action logs, monitoring for automation drift, protection against prompt injection, and an external pause or disable mechanism. After deployment, they should evaluate performance, review bias and error patterns, reassess controls when tools or permissions change, and safely retire agents that are no longer needed.
The guidance requires resourced checkpoints, clear ownership, audit logs, reversibility, and monitoring for both system drift and human overreliance.
🇨🇦 Canada’s Digital Regulators Set Out Common AI Principles
On 20 May 2026, the Canadian Digital Regulators Forum (CDRF) published a joint article on principles informing the development and use of artificial intelligence in Canada. The article was prepared by the four CDRF members: the Competition Bureau, the Office of the Privacy Commissioner of Canada (OPC), the Copyright Board of Canada, and the Canadian Radio-television and Telecommunications Commission (CRTC).
The article focuses on six principles: transparency, human agency and monitoring, human rights and democratic values, privacy and data governance, safety, security and robustness, and accountability. The CDRF notes that although these are often described as “principles,” some are already statutory requirements under laws administered by its members.
The Competition Bureau links AI transparency to deceptive marketing risk. It warns that AI-generated outputs may mislead consumers, shape preferences, or prevent informed choices, particularly through fake reviews, endorsements, impersonations, deepfakes, and tailored phishing.
The Copyright Board’s contribution highlights the use of copyrighted materials in training AI systems and the labeling or disclosure of AI-generated outputs. The article notes that Canada has not yet clarified whether AI developers require copyright owners’ permission for text and data mining activities. It also refers to recommendations from the House of Commons Standing Committee on Canadian Heritage calling for greater transparency from AI developers about copyrighted works used to train models, and for an opt-in system requiring creators’ express consent for training large language models.
Organizations developing, providing, or using generative AI must comply with applicable privacy laws, including PIPEDA principles such as consent, appropriate purposes, openness, accountability, limiting collection, limiting use and disclosure, and safeguards. The article also refers to the recent joint Canadian investigation into OpenAI’s ChatGPT, where regulators found that OpenAI’s collection of information for model training was overly broad and included sensitive personal information, including medical information and information about children.
The CRTC’s comments are relevant for telecommunications, broadcasting, and customer-facing AI tools. The article notes that AI may support network monitoring, predictive maintenance, and cyber resilience, but may also increase spam, scams, and other harmful communications. The CRTC also gives a practical example of accountability in AI-enabled call blocking: where AI is used to block fraudulent calls, there must be a review process, a contact point for complaints, and prompt correction of false positives.
🇺🇸 Vermont Enacts Law on Neurological Rights and AI in Health and Human Services
Vermont has enacted Act 101, a new law addressing neurological rights and the responsible use of artificial intelligence in health care, human services, and related public-sector contexts. The Act was signed by the Governor on 18 May 2026 and took effect on passage.
The law establishes a rights-based statutory foundation and directs Vermont’s Artificial Intelligence Advisory Council to prepare further recommendations by 15 January 2027. Those recommendations must address neurological rights, neurotechnology definitions, the use of generative AI by regulated professions, and the regulation of artificial and augmented intelligence in health insurance utilization review.
Act 101 recognizes that each individual has the right to mental and neural data privacy, freedom of thought, nondiscrimination in the development and application of neurotechnologies, and protection from unauthorized access to or manipulation of brain activity. The Act also recognizes protection from unauthorized neurotechnological alterations in mental functions critical to personality.
Vermont is treating neurotechnology and AI as systems capable of affecting autonomy, dignity, discrimination risk, and access to essential services.
The Act focuses on AI in health care and human services. The stated legislative intent is to promote ethical and responsible AI use while improving service delivery, coverage determinations, accessibility, quality of care, and population health outcomes. The Act refers to preventing harm, creating transparency, guaranteeing accountability, and minimizing the risks of artificial and augmented intelligence.
The Act also extends the sunset of Vermont’s Artificial Intelligence Advisory Council from 2027 to 2030 and expands the Council’s work. The Council must review guidance and recommendations from professional organizations, including those in medicine, social work, and education. It must also consider AI use in public participation and public finance and create opportunities for public education and engagement in AI policy development.
🇺🇸 NIST Reports 2025 Cybersecurity and Privacy Priorities
The U.S. National Institute of Standards and Technology (NIST) published NIST SP 800-238, Fiscal Year 2025 Annual Report for NIST Cybersecurity and Privacy Program in May 2026. The report covers NIST’s cybersecurity and privacy work during FY 2025, from 1 October 2024 to 30 September 2025, and summarizes activity across six main areas: cryptography, cybersecurity and AI, education and workforce, hardware and software security, infrastructure security, and risk management.
A central theme is the continued transition to post-quantum cryptography. NIST reports progress on PQC standardization, including the announcement of HQC as a fifth algorithm selected for standardization, work with industry and standards bodies to support adoption, and publication of a migration timeline. That timeline anticipates deprecation of quantum-vulnerable algorithms after 2030 and required use of quantum-resistant algorithms by 2035. NIST also reports expanded work through the Migration to PQC Project, involving more than 50 organizations, including demonstrations of discovery and inventory tools used to prioritize migration decisions.
For businesses, the first operational step is identifying where public-key cryptography is used across applications, cloud services, certificates, identity systems, VPNs, code signing, data-at-rest protections, third-party products, and long-lived sensitive records. Organizations with regulated, confidential, or long-retention data should also assess whether “harvest now, decrypt later” risk affects their threat model.
NIST highlights work on the Cybersecurity Framework Profile for AI, adversarial machine learning terminology, Dioptra as a testing platform for AI systems, and the Control Overlays for Securing AI Systems project. The latter aims to adapt NIST SP 800-53 controls and related AI materials for securing AI systems.
Organizations deploying AI systems should expect questions about model robustness, adversarial attacks, hallucination risk, dataset integrity, logging, access control, and supply-chain exposure. AI systems should be brought into existing risk registers, asset inventories, vendor review processes, and incident response plans.
The report also points to growing attention to secure software development and hardware trust. NIST reports the launch of the Secure Software Development, Security, and Operations Practices Project, an industry consortium, and a preliminary draft of SP 1800-44A. It also notes work on semiconductor security, supply-chain trust, provenance, and security measurement techniques.
For companies selling software, especially to enterprise or public-sector customers, this means that customers are likely to keep asking for evidence of secure SDLC practices, vulnerability management, patching processes, dependency tracking, SBOM-related practices, configuration baselines, and supplier controls.
NIST also reports several infrastructure-security outputs, including guidance on API protection for cloud-native systems, service mesh proxy models, the CSF 2.0 Manufacturing Profile, 5G cybersecurity papers, high-performance computing security guidance, IoT onboarding work, and water-sector cybersecurity collaboration. T
NIST reports continued adoption work for CSF 2.0, publication of SP 800-55 Volumes 1 and 2 on security measurement, updates to SP 800-53 and SP 800-53A for secure and reliable software updates and patches, publication of draft Privacy Framework 1.1, and continued work on OSCAL for security program automation. It also notes the release of Revision 4 of the Digital Identity Guidelines.
🇺🇸 NY DFS Warns Regulated Entities to Prepare for Heightened Cyber Threats Linked to Frontier AI
On 21 May 2026, the New York State Department of Financial Services (“DFS”) issued two related industry letters for entities regulated under DFS supervision. The first sets out measures that regulated entities should consider when operating in a heightened cybersecurity threat environment. The second addresses specific risks associated with frontier AI models, particularly their potential to increase the speed, scale, and effectiveness of vulnerability discovery and exploitation. DFS states that neither document creates new legal requirements, but both are intended to inform risk management and compliance under 23 NYCRR Part 500.
DFS defines a heightened threat environment as one where cybersecurity risks are significantly elevated and have a high likelihood of affecting information systems, nonpublic information, or operations. The Department gives two examples: geopolitical events that increase cyberattack risk, and technological developments that materially alter cybersecurity risk, including the release of frontier AI models.
The guidance frames Part 500 compliance as a baseline. DFS notes that regulated entities may need to take steps beyond minimum regulatory controls, depending on the threat, their information systems, supply-chain dependencies, and sector-specific risks.
The recommended measures are organized around three areas: reducing the attack surface, improving threat detection and readiness, and strengthening resilience and response. DFS points to practical controls such as rapid remediation of known exploited vulnerabilities, disabling unnecessary ports and protocols, stronger controls over MFA enrollment and changes, phishing-resistant MFA, network segmentation, privileged access reviews, validation of cloud configurations, secure programming practices, and restrictions on unsafe inputs before generating outputs or executing commands.
DFS also expects regulated entities to improve monitoring and operational readiness. The guidance refers to up-to-date intrusion prevention and detection controls, logging and security event alerting, use of threat intelligence and indicators of compromise, staff alerts about social engineering campaigns, monitoring of third-party code and permissions, and engagement with critical third-party service providers. On resilience, DFS highlights backup integrity, immutability and restorability, testing of recovery time objectives, review of incident response and business continuity plans, communications planning for prolonged disruption, and the ability of operational technology functions to continue if other information systems are compromised.
DFS warns that certain frontier AI models may amplify the potency, speed, and scale of identifying vulnerabilities and exploits. Although such models may not yet be broadly available, DFS urges CISOs of regulated entities to prepare now by improving security posture, reviewing risk assessments, and considering whether legacy or end-of-life systems should be replaced as part of operational resilience.
DFS places particular emphasis on expedited vulnerability management, third-party dependency mapping, secure programming practices, and stronger monitoring. Regulated entities should reassess vulnerability criticality and remediation timelines in light of the possibility that AI tools may make exploitation faster and easier. They should also coordinate with critical third-party and downstream providers, maintain dependency maps, validate third-party code and permissions, and clarify security responsibilities with providers.
The advisory also addresses AI-generated code and AI-assisted vulnerability remediation. DFS recommends additional testing, validation, and human oversight before AI-generated code is deployed in production. Where AI is used to identify or remediate vulnerabilities, organizations should guard against unknown changes in code or configurations and against inadvertent degradation of necessary code.
🇺🇸 Florida Supreme Court Sets Uniform Certification Rule for AI-Related Citation Accuracy
On 28 May 2026, the Supreme Court of Florida amended Florida Rule of General Practice and Judicial Administration 2.515(d)(2) to require every signer of a court filing to represent that “the legal authorities identified exist and are accurately cited.” The amendment takes effect on 15 June 2026 at 12:01 a.m. and applies to filings submitted by both attorneys and self-represented parties. The Court also authorized sanctions where a filing is inconsistent with that representation, after notice and an opportunity to be heard.
The change responds to the use of generative AI tools in legal drafting and research. The Court noted that large language models and similar systems may assist with preparing filings, but can also produce plausible-looking yet false content, including fabricated or hallucinated case law and legal citations.
Courts may impose sanctions for inconsistent filings. The listed sanctions include reprimand, contempt, striking the document, dismissal of proceedings, costs, attorneys’ fees, or other appropriate sanctions.
The Court’s commentary explains that the amendment is intended to create a statewide, uniform replacement for different circuit court administrative orders that had imposed varying disclosure and certification obligations concerning artificial intelligence in filings.
OECD Warns of Growing Fragmentation in Cybersecurity Regulation
The OECD has published Towards international coherence of cybersecurity regulations, OECD Digital Economy Papers No. 384, approved by the OECD Digital Policy Committee on 20 May 2026. The paper addresses a problem that cybersecurity obligations are expanding quickly, but they are not developing in a coherent way across jurisdictions, sectors, or regulators.
The report defines regulatory fragmentation as the existence of different, and sometimes conflicting, rules for the same or similar cybersecurity risks, services, products, or activities. In practice, this includes diverging national laws, inconsistent sectoral requirements, overlapping regulatory mandates, and the absence of shared definitions for basic concepts such as “cyber incident,” “critical infrastructure,” and “essential services.”
The OECD’s central point is that the multiplication of poorly aligned rules can weaken the very security outcomes that regulation is intended to improve. Organisations operating across borders must often interpret and implement several regimes at once, each with its own incident-reporting timelines, thresholds, documentation duties, audit expectations, and enforcement mechanisms. This creates additional legal and operational work that does not always translate into better protection.
The report identifies several causes of this fragmentation. Cybersecurity remains closely tied to national security and strategic autonomy, so governments are often reluctant to rely on external frameworks or international standards. Regulation also develops sector by sector, particularly in finance, energy, health, telecommunications, and digital infrastructure. This produces rules that may be justified within a sector but difficult to reconcile across a wider enterprise. The OECD also points to crisis-driven regulation, protectionist measures, and the growth of multiple public authorities with overlapping cyber, privacy, consumer protection, national security, and competition mandates.
The OECD notes that fragmented requirements increase compliance costs, duplicate documentation, and force organisations to spend scarce security resources on regulatory mapping rather than core cyber controls. This burden is especially acute for SMEs, which may face obligations similar to those imposed on larger entities but without dedicated compliance teams. The report cites NIS2 as one example of the scale of the challenge, noting expected impacts on more than 160,000 businesses in Europe and potentially many more globally through supply chains.
The report also highlights that inconsistent reporting and information-sharing rules can weaken cross-border co-operation. If incident definitions, thresholds, timelines, and reporting channels differ materially, public authorities may receive incomplete or non-comparable information. This can slow joint investigations, reduce situational awareness, and complicate crisis response. Fragmentation can also distort market incentives by encouraging companies to avoid jurisdictions with complex or duplicative requirements, rather than invest in stronger cybersecurity practices.
The OECD maps several existing efforts to improve coherence. At the domestic level, the United States has used instruments such as the NIST Cybersecurity Framework and federal cyber incident reporting reforms to reduce duplication. At the EU level, NIS2, DORA, the Cyber Resilience Act, the Cybersecurity Act, and the AI Act illustrate both the benefits and limits of regional coordination. These instruments create common baselines, but they also interact in complex ways. The OECD notes that the EU’s Digital Omnibus work seeks to simplify and align parts of the digital regulatory framework, including cyber incident and data breach reporting.
The paper also refers to bilateral and multilateral initiatives, including EU–U.S. work on cyber incident reporting alignment, digital partnerships, cyber dialogues, and possible mutual recognition mechanisms. Technical standards and certification schemes, including ISO/IEC 27000, the NIST Cybersecurity Framework, ITU recommendations, the EU cybersecurity certification framework, and the Common Criteria Recognition Arrangement, are presented as useful tools for alignment. However, the OECD cautions that standards remain non-binding and may be interpreted differently across jurisdictions.
🇻🇦Pope Leo XIV Issues Encyclical on Human Dignity in the Age of AI
On 15 May 2026, Pope Leo XIV issued Magnifica Humanitas, an encyclical on safeguarding the human person in the time of artificial intelligence.
The encyclical frames AI as part of the “new things” of the present age, comparable in social significance to the industrial transformations addressed by Rerum Novarum in 1891. The document accepts that technology can heal, connect, educate, and improve human life. The question is whether technological development is governed by the common good, human dignity, solidarity, and justice, or whether it becomes another instrument for concentration of power.
The encyclical warns against treating AI systems as equivalent to human persons. AI can process data, imitate language, produce analysis, and simulate empathy, but it does not have lived experience, conscience, moral responsibility, or relationships in the human sense.
AI is described as never being a purely technical matter when it affects rights, opportunities, status, and freedom. Responsibility should be defined across the full AI lifecycle: design, development, deployment, use, monitoring, challenge, and remedy. The encyclical is particularly concerned about opaque systems that make it difficult to identify who is accountable for harm, bias, exclusion, or error.
The encyclical also addresses the concentration of digital power. It notes that control over platforms, infrastructure, data, and computing capacity often rests not with states, but with large private technology actors. These actors may shape access to information, visibility, economic opportunity, and participation in public life. The document therefore treats data, algorithms, platforms, and technological infrastructure not only as private assets, but as goods with public consequences.
The encyclical warns that automation and AI may improve productivity while also de-skilling workers, increasing surveillance, and forcing people to adapt to machine-driven systems. It calls for technology to be designed around the human person, not merely around performance metrics. The document argues that companies should treat the quality and dignity of work as part of their measure of success. It also calls for retraining, worker participation, and social criteria for innovation when automation is introduced.
The encyclical points to the hidden human labor behind AI systems, including data labeling, model training, content moderation, resource extraction, and the work required to maintain digital infrastructure. It warns that apparently “immaterial” digital services may depend on exploitative labor, dangerous extraction conditions, and opaque global supply chains. Businesses using or developing AI should therefore view AI governance not only as a privacy, cybersecurity, or product compliance issue, but also as a human rights and procurement issue.
The document also links AI and digital platforms to freedom, addiction, and social control. It criticizes business models that monetize attention and exploit vulnerability, especially where minors are concerned. It warns that large-scale data collection and algorithmic profiling can create new forms of control by shaping what is visible, rewarded, suppressed, or penalized.
The encyclical further addresses disinformation and democratic communication. AI is described as an amplifier of manipulated content, misleading narratives, and synthetic media. The document calls for an “ecology of communication” based on verification, serious journalism, transparency in content-selection systems, protection of personal data, and education in the critical use of digital tools.
The encyclical takes a firm position against delegating lethal or irreversible decisions to artificial systems. It argues that moral judgment cannot be reduced to calculation and that chains of responsibility must remain identifiable and verifiable.
***
Direct your questions to groundcontrol@kepler.consulting.
Until the next transmission, stay secure and steady on course. Ground Control, out.