"AI Must Serve Humanity"- Pope Leo’s Big Warning on AI
++ OpenAI publishes third‑party evaluation playbook + frontier governance framework for safety, risk, security, and compliance; Anthropic launches Opus 4.8... & more
Today’s highlights:
Pope Leo XIV’s first encyclical, Magnifica Humanitas, has turned into a major AI-governance story because it frames artificial intelligence as a moral and social question, not just a technology question. Released on 25 May 2026, the document warns that AI could deepen inequality, replace or devalue human work, expand surveillance, damage the environment through energy- and water-heavy infrastructure, and make war easier by distancing humans from responsibility. The Pope’s core message is simple: AI must serve human dignity, justice, peace, truth, and the common good- not simply efficiency, profit, or power. Vatican News summarized the encyclical’s central warning as: AI must serve humanity and must not concentrate power.
The Anthropic angle is important. On 25 May 2026, Anthropic published a post titled “Anthropic co-founder Chris Olah’s remarks on Pope Leo XIV’s encyclical ‘Magnifica humanitas’.” In that post, Anthropic said Olah was invited to speak at the Vatican presentation as part of the company’s effort to “widen the conversation” around AI. Olah’s remarks were unusually self-critical for a frontier AI company representative: he said every major AI lab, including Anthropic, works under incentives that can conflict with “doing the right thing,” such as commercial pressure, geopolitical pressure, ambition, and the need to stay at the research frontier. He argued that AI questions are too big for computer scientists alone and need religion, philosophy, civil society, governments, and public critics. He highlighted three moral questions: how AI gains will be shared with the global poor, how humans can flourish in an AI-shaped world, and how society should think about increasingly mysterious AI models.
The controversy is that critics are asking whether Anthropic’s Vatican presence was genuine ethical engagement or reputation-building. The Guardian called this concern “Vatican-washing”- meaning Anthropic may benefit from appearing close to the Pope’s moral warning while still building the kind of powerful AI systems and infrastructure that the encyclical questions. The criticism is especially sharp because the Pope warned about job displacement, environmental harm, and concentration of power, while Anthropic is itself a leading frontier AI company expanding model capabilities and infrastructure. At the same time, the Guardian also noted that Anthropic and the Vatican do seem aligned on some issues, especially concern about AI in warfare and the need for external moral pressure on AI labs.
At the School of Responsible AI (SoRAI), we help both individuals and organizations build practical, real-world AI literacy and Responsible AI capability through structured, engaging, and action-oriented programs. For individuals, this includes AI Literacy, globally relevant certification training such as AIGP, RAI, and AAIA, as well as career transition and advisory support for professionals moving into AI governance roles. For organizations, we offer customized enterprise AI literacy training, Responsible AI strategy and governance setup, and AI assurance support to help teams understand, operationalize, and validate AI responsibly. At the core of SoRAI is a progressive three-layer approach: first helping people understand AI, then build the right governance foundations, and finally validate readiness through assurance and audit-focused thinking. Want to learn more? Explore our AI Literacy programs, certification trainings, and career support offerings, or write to us for customized enterprise solutions.
⚖️ AI Ethics
Universal Music Group and TikTok Renew Licensing Deal to Remove Unauthorized AI-Generated Music
Universal Music Group and TikTok have renewed their licensing deal, with both companies agreeing to remove unauthorized AI-generated music and improve crediting for artists and songwriters. The agreement marks a reset after a 2024 dispute that led UMG to pull its catalog from TikTok over concerns about AI-generated tracks and copyright enforcement. The move comes as the music industry faces growing pressure from fake songs and voice-cloning tools that can mimic major artists and spread quickly online. The deal could also influence how other tech platforms handle AI, copyright, and creator protections as regulatory scrutiny increases.
YouTube to Automatically Label AI-Generated Videos and Make Disclosures More Prominent Across Platform
YouTube will begin automatically labeling videos when its systems detect significant photorealistic AI, instead of relying only on creators to disclose that content themselves. The platform said its core AI labeling policy is unchanged, but starting in May it will use internal signals, along with C2PA metadata and checks tied to its own AI tools, to add labels when needed. The labels will also become more visible, appearing below long-form videos and directly on Shorts, while less realistic or lightly altered AI content may still be labeled only in the description. YouTube said these labels will not affect recommendations or monetization, and the move comes as more advanced AI video tools make realistic synthetic content harder to identify.
Meta CEO Mark Zuckerberg Promises No More Company-Wide Layoffs This Year After 8,000 Job Cuts
Meta CEO Mark Zuckerberg told employees that there will be no further company-wide layoffs this year after recent restructuring led to about 8,000 job cuts. The layoffs were tied to Meta’s broader effort to streamline operations while spending more on artificial intelligence. Zuckerberg also said the company had not handled communication well during the process and promised to be clearer with staff going forward. The message was aimed at reassuring employees as Meta continues to adjust its business and priorities.
Study Finds Nearly 1.46 Lakh AI-Hallucinated References Entered Scientific Papers in 2025
A new study says nearly 1.46 lakh AI-hallucinated references entered scientific papers in 2025, highlighting a growing problem in academic publishing. The report found that many fabricated citations are slipping past peer review and appearing in published journals, raising fresh concerns about the reliability of scientific records. It said the issue is affecting less experienced researchers and solo authors more heavily, as they may rely more on AI tools without strong verification. The findings also suggest that current safeguards used by journals and publishers are not strong enough to catch these fake references before publication.
Google Search AI Overviews Struggle to Define ‘Disregard,’ ‘Stop,’ and ‘Ignore’ Amid Rollout Issues
Google’s AI-powered Search is misfiring on simple dictionary queries such as “disregard,” “stop,” and “ignore,” sometimes showing an AI Overview and blank space instead of a standard definition snippet. The issue has been widely reproduced by users and media reports, although some searches still return the correct definition in certain cases, including in Incognito Mode. Google said it is aware that AI Overviews are misreading some action-related queries and that a fix is being rolled out soon. The glitch is less severe than some earlier AI Overview errors, but it adds to concerns about how smoothly Google’s shift from traditional search results to AI-generated answers is working.
OpenAI Publishes Shared Playbook for Trustworthy Third-Party Frontier AI Evaluations and Reporting Standards
OpenAI has published a playbook for third-party evaluations of frontier AI models, arguing that trustworthy testing now depends not just on prompts and scores, but also on the “harness” around a model, including its tools, memory, retries, and workflow setup. The company says evaluation reports should clearly state whether they are testing capability, safeguard strength, or comparing models, and should disclose the budget, environment, and elicitation methods used. It also warns that results can be distorted by reward hacking, refusals, contamination, broken tasks, or deliberate underperformance, and says these risks should be checked and reported. The guidance is aimed at shaping stronger industry standards as frontier models become more agentic and harder to assess with older chatbot-style benchmarks.
OpenAI Publishes Frontier Governance Framework to Align Safety Practices With Emerging AI Regulations
OpenAI has published its Frontier Governance Framework, a public document outlining how the company’s safety and security practices align with emerging AI rules, including California’s Transparency in Frontier AI Act and the EU AI Act’s Code of Practice for general-purpose AI. The framework builds on OpenAI’s existing Preparedness Framework, which remains its main system for managing major risks from advanced AI models. It covers risk assessment and mitigation in areas such as cyber abuse, CBRN threats, harmful manipulation, and loss of control, along with model reporting, security management, incident response, outside expert input, and future updates. OpenAI said the framework is expected to evolve as model capabilities, testing methods, and regulatory requirements continue to change.
Sam Altman Says AI Jobs Apocalypse Probably Will Not Happen as Adoption Lags
OpenAI CEO Sam Altman said he no longer expects the “jobs apocalypse” from AI that he once warned about, saying the impact on entry-level white-collar work has been slower and smaller than he anticipated. He said he had underestimated how much the human side of work matters, arguing that people still value interacting with other people on the job. The shift comes as businesses continue to question whether AI delivers enough value to justify its high costs, even as adoption keeps growing. Economists remain divided, with some studies finding little evidence so far of major AI-driven unemployment, while layoffs tied to AI investment and efficiency pushes are still rising across the tech sector.
🚀 AI Breakthroughs
Anthropic Secures $65 Billion Series H Round at $965 Billion Post-Money Valuation
Anthropic said it has raised $65 billion in a Series H round at a reported $965 billion post-money valuation, with backing from major investment firms and strategic chip partners. The company said demand for its Claude AI models has surged across enterprises worldwide, with annualized revenue passing $47 billion earlier this month. Anthropic said the new funding will be used to expand computing capacity, support safety and interpretability research, and scale products and partnerships tied to Claude. It also said it has secured major infrastructure agreements with Amazon, Google, Broadcom, and SpaceX as it works to meet rising demand for AI services.
Anthropic Releases Opus 4.8 and Dynamic Workflows Amid Pressure From OpenAI and Google
Anthropic has released Opus 4.8, the latest version of its most advanced public AI model, just 41 days after Opus 4.7, signaling a faster-than-usual update cycle as competition from OpenAI and Google intensifies. The company said the model keeps the same pricing as the previous Opus release and improves how it handles uncertain or flawed data, with testers saying it is more likely to flag doubts and unsupported claims. Alongside the model, Anthropic also opened a research preview of Dynamic Workflows, a tool designed to let larger models coordinate complex jobs through hundreds of parallel subagents. The company said this could help Claude Code handle large codebase migrations across hundreds of thousands of lines of code. Anthropic also said its more advanced Mythos model remains on hold over cybersecurity safeguards, but it expects to make Mythos-class systems available in the coming weeks once protections are ready.
YouTube Adds AI Podcast Recommendations, Auto Speed, and On-the-Go Listening Features for Premium Users
YouTube has added several new podcast features for Premium users as it steps up efforts to compete with Spotify, Apple Podcasts, and video podcast rivals such as Netflix. The update includes an AI-powered recommendation tool that suggests podcasts based on genre, mood, or shows a user already likes, along with a new “Auto speed” setting that adjusts playback speed during slower or denser parts of a conversation. The company has also added an on-the-go listening mode with simple controls for skipping, rewinding, and moving to the next episode while using background playback. Auto speed and on-the-go mode are available now on Android for Premium users, with iOS support expected in the coming months.
Deepseek Makes 75 Percent Price Cut Permanent, Undercutting GPT-5.5 Output Token Costs by 34x
DeepSeek has made the 75% discount on its flagship DeepSeek V4 Pro permanent, keeping prices at $0.435 per million input tokens and $0.87 per million output tokens after the offer was due to end on May 31, 2026. At those rates, the model is about 11.5 times cheaper than GPT-5.5 on input and roughly 34.5 times cheaper on output, with an even wider gap against long-context pricing and Anthropic’s Opus 4.7. DeepSeek’s V4 Pro and V4 Flash also offer a 1 million-token context window, support up to 384,000 output tokens, and work with both OpenAI- and Anthropic-style APIs, making switching easier for developers. The move sharpens China’s AI pricing pressure on Western labs, especially as companies facing uncertain AI returns increasingly look for lower-cost models that are good enough rather than the best-performing ones.
Microsoft Redesigns 365 Copilot With Faster Performance, Task-Aware Workspace, and Unified App Experience
Microsoft has redesigned the Microsoft 365 Copilot app and its in-app experience across Word, Excel, PowerPoint, and Outlook to make the AI assistant faster, cleaner, and more task-aware. The update expands the prompt area into a workspace, adds a unified entry point across Microsoft 365 apps, and brings context-based actions and side-pane editing tools that can work directly inside documents, slides, cells, and paragraphs. Microsoft said the Copilot app now loads more than twice as fast, with load times cut by over 50%, while response times for complex chat prompts improved by 10%. The company also said usage has risen after the rollout of the new in-app designs, increasing by 27% in Word, 33% in Excel, 43% in PowerPoint, and 30% in Outlook, as it shifts Copilot from a basic chat interface toward a more connected and agent-like work assistant.
Leaked Siri App Preview Shows Apple Preparing AI Upgrade to Challenge ChatGPT on iPhone
Just ahead of WWDC, a report based on leaked renders said Apple is preparing a major Siri overhaul in iOS 27, including a new standalone Siri app aimed at competing with ChatGPT, Gemini, and other AI assistants. The upgraded Siri is expected to appear through the Dynamic Island for quick voice responses, while swipe-down search may be rebuilt around AI-powered Siri for tasks such as finding information, opening apps, sending messages, and managing notes or calendars. The report said Apple is using Google’s Gemini technology to strengthen Siri’s intelligence while also developing its own on-device AI features focused on privacy. The new Siri app is also expected to support chat history, text conversations, and uploads of documents and photos, showing Apple’s broader push to bring AI tools to its large device base.
Trump Appoints Former Attorney General Pam Bondi to White House Artificial Intelligence Advisory Panel
US President Donald Trump has appointed former Attorney General Pam Bondi to a White House advisory panel on artificial intelligence, according to a report by Axios. The White House has not officially confirmed the appointment. The reported move suggests Trump is continuing efforts to influence how the US approaches AI policy and regulation. Bondi’s selection also points to the administration’s interest in placing trusted political allies in key technology advisory roles.
🎓AI Academia
Researchers Define Agentic AI Technical Debt and Stochastic Tax for Enterprise Governance Systems
A new paper argues that agentic AI systems create a distinct form of technical debt because they do more than generate outputs: they plan, call tools, use memory, and take actions across workflows. The authors describe this as “Agentic Technical Debt,” or the governance and design risk that builds up when prompts, tool connections, memory, and control rules are assembled faster than they can be properly validated and audited. The paper also defines “Stochastic Tax” as the ongoing operational cost of keeping these probabilistic systems within safe and acceptable limits through monitoring, retries, guardrails, and human oversight. Its central message is that companies deploying agentic AI need lightweight dashboards and governance controls to track both the long-term liability and the recurring cost of managing unpredictable behavior.
Study Proposes Unified Social Media Cyberbullying Governance Framework From Content Detection to Intervention
A new survey paper from researchers at Beijing University of Posts and Telecommunications argues that cyberbullying governance on social media needs to move beyond one-off post detection and toward a full lifecycle approach. The study organizes existing research into four linked stages: identifying harmful content, modeling user behavior, tracking how toxic events spread for early warning, and designing interventions to reduce harm. It also reviews datasets and evaluation methods used in the field, while highlighting key challenges such as multimodal content, explainability, fairness, and the dual-use risks tied to generative AI. The paper frames cyberbullying as a broader behavioral and networked problem, not just a content moderation task, and calls for more proactive systems to build safer online platforms.
OpenAI Details Frontier Governance Framework for Systemic Risk Mitigation, Security Management, and Regulatory Compliance
OpenAI has published a Frontier Governance Framework outlining how it plans to assess and reduce systemic risks from its most advanced AI models. The document says the framework is designed to align with legal requirements under California’s Transparency in Frontier AI Act and the European Union’s AI Act, while covering risks such as cyber offense, CBRN threats, harmful manipulation, and loss of control. It also details processes for safety mitigations, critical incident detection and response, security risk management, model reporting, and external expert input. The company says the framework applies to certain deployed frontier models and will be updated over time as its practices and regulatory obligations evolve.
About SoRAI: SoRAI is committed to advancing AI literacy through practical, accessible, and high-quality education. Our programs emphasize responsible AI use, equipping learners with the skills to anticipate and mitigate risks effectively. Our flagship AIGP certification courses, built on real-world experience, drive AI governance education with innovative, human-centric approaches, laying the foundation for quantifying AI governance literacy. Subscribe to our free newsletter to stay ahead of the AI Governance curve.
