OpenAI and Google answered AI week in the language of agents. Anthropic answered it, at least publicly, in the language of moral responsibility. OpenAI's news page and its May 22 Gartner announcement present Codex as an enterprise coding agent, used by more than 4 million people each week and sold through governance, sandboxing, approval gates and deployment controls. Google's AI pages and Gemini API post describe managed agents that run in secure cloud sandboxes and can be defined as versionable files with AGENTS.md and SKILL.md. Anthropic's own newsroom, meanwhile, published Chris Olah's remarks on Pope Leo XIV's encyclical, "Magnifica humanitas." [1] [2] [3] [4] [5]

Tuesday's paper covered Pope Leo's AI encyclical with Olah on the Vatican dais. It also covered Anthropic's acquisition and Microsoft's FARA push as pressure on OpenAI's operator model. Today's follow-up is not that OpenAI or Google went silent. They did not. They spoke loudly. They simply spoke in the grammar of enterprise control while Anthropic's public artifact sat inside a papal moral conversation.

OpenAI's Gartner post is almost a miniature constitution for the new software firm. Codex, it says, understands large codebases, uses tools, makes changes, runs tests and prepares work for review. For enterprises, the pitch is "speed with control," not wonder. The quoted sales language says companies are no longer asking only whether AI can write code; they are asking how to deploy agentic systems safely at scale as a new operating layer. [2]

Google's post is more architectural and, therefore, more revealing. Managed Agents in the Gemini API can spin up an agent in an isolated Linux environment, use tools, execute code and be defined through files. [4] This is not the chatbot era with a nicer robe. It is platform infrastructure. The agent becomes something closer to a worker process with permissions, logs and deployable instructions.

That makes the Vatican absence more interesting, not less. The moral question around frontier AI is often discussed as if companies must choose between ethics and product. The public record this week shows a subtler division. Anthropic chose to attach one of its founders to an encyclical release. OpenAI and Google chose to attach their names to managed execution, governance controls, coding agents and developer infrastructure. [2] [4] [5]

X could not be quoted responsibly here. Three searches for real status posts on OpenAI, Google, managed agents and Magnifica returned no verified /status/ URL. The article therefore does not pretend to import platform chatter. The divergence is visible in the documents themselves. The public moral frame is not absent from the industry; it is unevenly inhabited.

The word "Magnifica" can sound like ornament. It is not, at least not for a company that wants to present itself as safer, slower, more interpretable and less captured by the spectacle of capability. Anthropic benefits when the moral conversation has a named seat and its competitors are represented, in the public stack, by product posts. That benefit may be branding. It may be conviction. In public life, the difference is often tested only when money arrives.

OpenAI's post has its own morality, just not the Vatican's. Approval gates, RBAC, sandboxing and auditable governance are ethical claims in technical clothing. [2] Google makes a related claim when it emphasizes isolated cloud sandboxes and versionable agent files. [4] The companies are saying: trust us because the agent can be controlled. The Vatican frame asks a prior question: controlled toward what?

That is why this belongs in technology rather than theology. The operational story is that the agent is becoming an enterprise surface. The political story is that the agent's builders are choosing where to place their public vocabulary. One vocabulary says human dignity. Another says governance. Another says sandbox. The market will reward whichever one can be bought without embarrassment.

The next receipt to watch is not a speech. It is a policy, filing, customer deployment, safety commitment or product constraint that links these vocabularies instead of letting them live apart. If OpenAI and Google keep selling agentic control while Anthropic keeps owning the moral dais, the divergence will become part of the competitive landscape. If they answer the Vatican frame directly, the paper should ask what they changed beyond the paragraph.

-- ANNA WEBER, Berlin