The Pope Enters the Room — and He Comes With 42,000 Words
The most talked-about AI governance development this week did not come from a legislature or a regulatory body. On May 25, Pope Leo XIV released his first major papal teaching document, a 42,300-word encyclical titled Magnifica Humanitas, framing AI as ushering in a “new industrial revolution” and calling for stronger government and corporate regulation of the technology.
The document carries unusual weight — and unusual symbolism. Leo chose his papal name in explicit reference to Pope Leo XIII, whose 1891 encyclical Rerum Novarum addressed the social consequences of the Industrial Revolution and became the foundation of modern Catholic social teaching on labor and technology. The parallel is deliberate and pointed: the Church is positioning AI not as a technical curiosity but as a civilizational inflection point requiring moral guardrails.
The encyclical targets not any particular type of AI, but rather the broader effects of technologies that “merely imitate certain functions of human intelligence.” It addresses war, wealth inequality, modern slavery, and the erosion of democracy as connected consequences of unchecked technological power.
What made the Vatican event particularly notable was the guest list. The presentation on May 25 featured testimony from the pope himself, prominent cardinals and theologians, and Christopher Olah, a co-founder of the AI company Anthropic who leads its interpretability team. Anthropic’s interpretability work — focused on understanding what is actually happening inside AI systems — sits at the technical core of the safety argument the Church is making at a moral level. That convergence on a single stage is not a coincidence.
The encyclical specifically flags the risks of uncritical reliance on AI, its environmental impact, its role in entrenching the power of dominant private actors, and the need for constraints on the use of AI in warfare.
Illinois Passes the Nation’s Most Demanding Frontier AI Safety Bill
The Illinois legislature sent a landmark AI accountability bill to Governor Pritzker this week, and it is materially more demanding than anything currently on the books in the United States.
Senate Bill 315 would require frontier AI companies to create, publish, and annually update plans addressing severe or catastrophic risks from their AI models. It would also mandate annual independent third-party audits of AI companies on safety issues — a first for any AI legislation in the US.
The bill passed unanimously in the House and was modeled after 2025 laws in New York and California, with sponsors explicitly citing the absence of a federal standard as justification for state-level action. The thresholds are calibrated to target the largest players: SB 315 applies to companies with revenues exceeding $500 million and significant computing capacity.
OpenAI and Anthropic both publicly supported the bill, while a trade organization representing other AI companies opposed it. Google, xAI, and Meta did not respond to comment requests. Governor Pritzker has indicated he intends to sign it.
Beyond audits and safety plans, SB 315 would create whistleblower protections and internal reporting processes for employees at AI companies. That last provision is significant — it creates a legal channel for AI safety researchers who observe problems to report them without career risk. No comparable protection currently exists at the federal level.
Colorado Retreats — A Cautionary Tale About Legislative Ambition
Not all movement this week was forward. Colorado’s trajectory is worth examining as a case study in how political pressure can erode regulatory ambition.
On May 14, Governor Jared Polis signed SB 26-189, which repeals and replaces the original SB 24-205 entirely. The replacement was the product of a governor-appointed AI policy work group. The original bill was one of the most comprehensive state AI laws in the country, requiring proactive algorithmic discrimination prevention, risk management programs, NIST alignment, annual impact assessments, and self-reporting requirements.
Senate Majority Leader Robert Rodriguez, the author of the original law, described the replacement as “more of a notice bill” that retains the core principle of disclosure when AI is used in consequential decisions. The duty-of-care standard, mandatory NIST alignment, annual impact assessments, and algorithmic discrimination self-reporting requirements of the original law were all removed.
The rollback reflects a pattern that will likely repeat in other states: ambitious legislation introduced, industry pressure applied, and the final product narrowed to transparency and notice requirements that cost companies relatively little. Illinois took the opposite approach this week. The contrast between the two states will be instructive to watch as enforcement begins.
The EU AI Omnibus: A Deal, Finally — With Important Delays Baked In
The most consequential regulatory development of the month arrived earlier in May but continues to reverberate. In the early hours of May 7, after roughly nine hours of negotiations and a previous trilogue that had collapsed on April 28, the European Parliament, Council, and Commission reached political agreement on the AI Omnibus.
The deal’s most significant practical effect is a deadline extension. The agreement postpones the high-risk obligations to December 2, 2027 for Annex III systems — covering biometrics, employment, education, law enforcement, critical infrastructure, and border management — and August 2, 2028 for Annex I systems, meaning AI embedded in products covered by EU sectoral safety legislation.
The deal also adds a new prohibition on AI systems generating child sexual abuse material or non-consensual intimate imagery — extending liability not just to intentional deployment, but to providers of general-purpose generative AI tools that fail to implement reasonable safeguards against such misuse.
The deal still requires formal adoption and publication in the EU Official Journal before August 2, 2026 to take legal effect. That clock is running.
The US Federal-State Divide: A Compliance Minefield
The broader US picture remains one of structural tension between a federal government opposed to binding AI regulation and states that are filling the vacuum aggressively.
The Trump administration tasked the Attorney General with coordinating litigation efforts against state measures that impose what the administration characterizes as innovation-limiting requirements, and in January 2026 announced a formal AI litigation task force for this purpose.
Several major state AI statutes are already in effect or scheduled to take effect imminently, including California’s AI Transparency Act, Texas’s Responsible Artificial Intelligence Governance Act, and Colorado’s replacement legislation, effective June 30, 2026. For multinationals, the divergence between US and EU postures creates a compliance reality driven by the stricter jurisdiction. The EU is enforcing binding rules with penalties up to 3% of global annual revenue. The US federal government is explicitly choosing not to act.
That asymmetry is not a theoretical concern — it is an operational reality for any company operating across both markets.
The Frame That Matters This Week
The through-line across all of this is a single structural fact: the entities most capable of shaping AI governance are not aligned. The Vatican is calling for a pause in the pace of development. Illinois is demanding third-party audits. Colorado retreated under industry pressure. The EU secured a deal but extended its own deadlines. The US federal government is actively working to preempt state regulation while offering nothing binding in its place.
What is absent is coherent, enforceable international coordination — and every week that passes without it is a week in which technical capabilities outpace governance frameworks. That gap is the story, and it is widening.
0 Comments