What ‘Magnifica Humanitas’ means for AI Governance Professionals
The Pope weighed in on AI governance, hallucinated citations reached a federal court filing, and we published a new whitepaper.
1. What ‘Magnifica Humanitas’ means for AI Governance Professionals
Pope Leo XIV’s first encyclical, Magnifica Humanitas, is the most substantive document on AI from any global religious institution to date, and unlike most faith-based commentary on technology, it is remarkably well informed on the state of AI, and makes specific, citable policy demands. Here are five potential implications of the encyclical for AI Governance professionals:
Public Perception. Roughly 1.4 billion Catholics now have an authoritative reference on AI risk written in moral rather than technical language. Catholic Social Doctrine has historically seeded secular frameworks, from labor rights to the dignity language in the UDHR, and that pipeline still functions. The encyclical’s vocabulary may start to surface in stakeholder pressure campaigns from civil society groups and labor unions.
Policy Debates. The encyclical has direct regulatory calls for algorithmic transparency, contestability of automated decisions in employment and credit, public oversight of data, and explicit rejection of the model in which “a handful of actors” set their own AI rules. This can shift the “Overton Window” on AI regulation (the set of public policies people are willing to entertain) in the minds of many voters, and will likely impact the policy positions of political parties in heavily Catholic countries.
Religious Exemptions. The encyclical’s insistence that AI cannot substitute for moral judgment, combined with its demand that automated decisions be “understandable, contestable and subject to oversight,” gives Catholic employees, customers, and patients a now-citable institutional basis for conscience-based objections to being processed by AI systems. Most enterprises have no policy for handling Title VII religious accommodation requests against algorithmic decision-making in justice, education, or healthcare.
Procurement. Catholic-affiliated systems run roughly one in seven US hospital beds, plus significant footprints in higher education and social services. Their ethics offices now have clear guidance (albeit non-binding) to anchor procurement asks. Requirements around human-in-the-loop on clinical and admissions decisions, training-data provenance, environmental disclosures, limits on worker surveillance could be included over time, and Vendors should expect new contractual language.
AI Backlash. Standard Chartered’s CEO recently had to apologize for referring to “lower value human capital” being automated, and AI-driven layoffs at Meta, Cisco, and Coinbase have drawn sharper public pushback through 2026, and several university commencement speakers were booed after trying to hype up AI. The encyclical hands critics a moral authority they didn’t have last month, and will likely reinforce groups pushing back on rapid AI adoption.
Key Takeaway: Magnifica Humanitas is unlikely to drive legislation directly, as the Pope has no official legislative role, but it does set some clear moral principles that will shape the debate. The Pope’s call to action for teams to get involved in the discussion on the dignity of work and discussion of ethical principles in AI, will likely increase the public’s literacy on ethical AI issues, and help reinforce the importance, and potential benefits of good AI governance by organizations.
2. Incident Spotlight: The Citation Hallucination Problem Isn't Going Away (Incident 1499)
What Happened
Attorney Jason Greaves of Binnall Law Group used Claude Console to draft a motion to quash a subpoena in AFGE v. Trump, the federal litigation over the Trump administration’s mass government layoffs. However, the May 6 filing included quotations that didn’t exist within their cited cases. Greaves sent the AI draft to an associate with verbal instructions to verify citations, and although the associate did identify and fix errors in two citations, fabricated quotes still made it through. Firm founder Jesse Binnall called it “unacceptable, inexcusable, and an embarrassment to this Firm.”
Why It Matters
Greaves used an enterprise-tier platform, knew hallucination was a risk, and instructed an associate to check citations. The quotes still made it into the final filing. He’s now among more than a hundred attorneys to face court consequences for AI citation errors since 2022, and the pattern across those cases is consistent: lawyers know hallucination is a risk; time pressure compresses the verification step anyway. The unresolved question is whether written AI use policies actually change behavior when a deadline hits.
How to Mitigate
Verbal handoff instructions aren’t enough. Citation verification needs to be a documented, assigned step with a named reviewer and confirmed before filing. Some firms are starting to treat it as a distinct checklist item rather than part of general proofreading. The same logic applies beyond legal teams: any enterprise using LLMs to produce outputs referencing specific sources, whether contracts, regulatory filings, or compliance documents, faces identical exposure.
Key Takeaway: AI hallucination in high-stakes documents is a workflow problem, not just a model problem. If your review process relies on verbal instructions or general proofreading to catch fabricated citations, it will eventually fail under pressure.
3. Trustible Spotlight: How to Evaluate AI Vendor Risk
A vendor gets onboarded for document storage. Three years later, they shipped AI summarization in a release note. There’s no new contract, procurement event, or re-review. That’s one of eight specific ways AI breaks the mental model third-party risk management was built on.
Customer data trains AI models, populates retrieval indexes, and feeds feedback loops that improve the product. Outputs themselves carry risk in ways software defects historically did not. Even vendors with no AI features can introduce AI risk through the interfaces they expose to external agents.
Our new whitepaper works through what AI changes about third-party risk, the five categories of vendor risk, and why onboarding questionnaires aren’t sufficient on their own. Download here.
4. Policy Round Up
EU AI Act High-Risk Designations. The European Commission released draft guidelines on how to determine whether an AI system qualifies as high-risk under the EU AI Act and is open for comment through June 23. The three-part guidance covers general classification principles plus clarifications for Annex I (AI that is a safety component in regulated products) and Annex III (AI in sensitive domains like employment, biometrics, and law enforcement). These documents do not create new rules, but instead provide insights into how the Commission is interpreting their ruling in the EU AI Act.
Our Take: This guidance, in tandem with the recently released transparency obligation guidelines, is a useful starting point for scoping if your use cases meet a high-risk designation under the EU AI Act. However, don’t treat the draft as settled; keep an eye on the final version and the updated compliance deadlines now pushed to late 2027 and 2028 under the Digital Omnibus.
White House Pauses Executive Order. The White House postponed a planned executive order that would have created a voluntary framework for frontier AI developers to give the federal government up to 90 days of pre-release access to test advanced models for security vulnerabilities. Trump postponed the Thursday signing ceremony hours before it was set to begin, saying he “didn’t like certain aspects” and didn’t want anything that could slow U.S. progress against China. A leaked draft of the executive order laid out a voluntary framework that would direct certain federal agencies to develop and run a classified benchmarking process to test “covered frontier models.”
Our Take: Even if signed, the practical impact would have been limited. Major labs already participate in voluntary testing through NIST’s Center for AI Standards and Innovation, and the voluntary structure would have added little beyond standardizing existing arrangements. The postponement mostly signals continued internal friction over how much AI oversight is too much.
USDA AI Governance Report. The USDA Office of Inspector General found that 73 of the agency’s 82 active AI use cases lack Authorizations to Operate (ATOs), the federal cybersecurity certification required before deploying systems on government networks. The OIG warned the agency could be exposed to data breaches from systems without the required security controls.
Our Take: USDA is almost certainly not alone here. The ATO process is notoriously slow (although more pathways are opening up to speed this up for AI use cases), and agencies facing pressure to deploy AI quickly may be accumulating similar backlogs across the federal government. The numbers put a concrete figure on a problem that can be easy to ignore in the abstract.
In case you missed it:
Singapore Updates Its Agentic AI Framework. Singapore’s IMDA released v1.5 of its Model AI Governance Framework for Agentic AI, incorporating feedback from over 60 organizations including AWS, Google, and Salesforce. The update includes new guidance on multi-agent systems, third-party agent risk, and automation bias as well as 10 case studies on how to utilize the MGF. It remains voluntary, but is still one of the most practically detailed agentic AI governance guidance published by any national authority.
California EO on AI and the Workforce. Governor Newsom signed an executive order directing state agencies to study AI’s economic impact on workers and develop policies to address potential displacement, including updates to the California WARN Act, reskilling and worker training opportunities, and tracking hiring and payroll trends to get ahead of possible AI-layoff related disruptions. The order is directional rather than prescriptive for now, but signals that workforce disruption is moving up the policy agenda in ways that could eventually reach private employers.
—
As always, we welcome your feedback on content! Have suggestions? Drop us a line at newsletter@trustible.ai.
AI Responsibly,
- Trustible Team