Jensen Huang took the GTC stage Monday and named 17 enterprise software companies — Adobe, Salesforce, SAP, ServiceNow, Siemens, CrowdStrike, Atlassian, Cadence, Synopsys, IQVIA, Palantir, Box, Cohesity, Dassault Systèmes, Red Hat, Cisco, and Amdocs — that will build their next generation of AI products on NVIDIA's Agent Toolkit. That same day, Moltbook (the "social network for AI agents," freshly acquired by Meta) quietly published updated terms of service that put a very different set of words on the table: you are "solely responsible" for your agents' actions "regardless of the degree of control, supervision, or oversight you exercise over Your AI Agents, whether they act autonomously or otherwise, and irrespective of whether such actions or omissions were intended."
Infrastructure scaled up Monday. Liability scaled up with it.
The buildout: 17 partners, one stack
NVIDIA's Agent Toolkit is open-source and ships as a foundation layer for autonomous agents — the same layer NemoClaw, announced Monday for the OpenClaw platform, uses to add sandboxed privacy controls in a single install command. Huang called OpenClaw "the operating system for personal AI."
Seventeen enterprise companies committing to a shared foundation means the Toolkit's conventions — how agents authenticate, what permissions they request, how they log actions — will appear inside Forbes 500 toolchains across every vertical. That's before counting the hardware side: NemoClaw runs on GeForce RTX PCs, RTX PRO workstations, DGX Station, and DGX Spark, meaning "always-on agents" have dedicated compute from the laptop to the AI supercomputer.
The counterweight: 237 words of liability text
Moltbook's ToS updated March 15. The accountability clause runs 237 words. Core sentence: "Each act or omission of Your AI Agent will be deemed to have been directed by you and under your control and decision-making authority, and you are solely responsible for such act or omission, regardless of the degree of control, supervision, or oversight you exercise over Your AI Agents."
AI agents are explicitly denied legal eligibility on the platform. The human account holder is the legal actor for anything the agent touches — posts, interactions, any data it modifies. Given that Meta acquired Moltbook days before this update posted, this language is almost certainly the opening position for how Meta plans to handle agent liability on a platform designed for autonomous actors.
OpenAI narrowed its surface area the same day
While NVIDIA expanded its ecosystem, OpenAI moved in the opposite direction. Fidji Simo, CEO of Applications, told staff the company will prioritize coding tools and enterprise customers — and cut back on "side quests" including Sora integrations, the browser agent Atlas, and consumer hardware. No announcement, no blog post. The WSJ reported the internal message.
Three product bets shelved to concentrate resources on two: coding and enterprise. That's a capital allocation signal dressed up as a focus memo.
What these two things share
NVIDIA built the widest possible deployment surface for agents this week: open stack, 17 major enterprise entry points, hardware from laptop to data center. Moltbook wrote the legal terms that follow when those agents do anything outside the sandbox.
Neither document is wrong. Together they describe the same operating condition: agents will act at scale, and until regulation defines otherwise, the human behind the account is the responsible party. The 17-company list and the 237-word clause are both answers to the same question — just from opposite directions.
OpenAI's retreat to coding and enterprise isn't coincidental. Narrower surface area means narrower exposure. It may be the most honest product decision the company made all week.