What changed
NVIDIA announced the Open Secure AI Alliance on 27 July 2026, naming agent harness research among its contributions. That research has a public repository: NOOA, which The Hacker News calls an Apache 2.0 research framework designed to make agent behavior easier to test, trace, audit, and govern. NVIDIA’s page says the project “is now available on GitHub”, which reads like a first release. It is not one: the NOOA paper went up on arXiv on 22 July 2026 and already links nvidia-nemo/labs-OO-Agents, so that sentence re-announces a repository the paper cited five days earlier.
An agent is a Python object
The paper describes a programming model that folds prompt templates, tool schemas and workflow graphs into one class: methods are the actions, fields are state, docstrings are prompts, annotations are contracts, and a method whose body is ... is completed at runtime by an LLM-driven agent loop. Normal bodies stay deterministic Python.
class SupportAgent(Agent):
"""You are a support agent for a customer service system."""
order_db: OrderDB
def is_refund_eligible(self, order: Order) -> bool:
"""Return whether an order is eligible for a refund."""
return order.delivered and order.days_since_delivery <= 30
@strategy(PredictStrategy())
async def classify(self, message: str) -> TicketKind:
"""Classify the customer message into the best ticket kind."""
...
@strategy(CodeActStrategy())
async def triage(self, message: str, photo: Image | None, order: Order | None) -> Ticket:
"""Triage a customer message and create a support ticket."""
...
Read that listing as a diff rather than as documentation. is_refund_eligible is ordinary code your tests already cover; classify and triage are prompts wearing a method signature. I think that is where the value and the exposure both sit: prompt text lands in pull requests.
The alliance without the model labs
The Hacker News records OpenAI, Google and Meta as absent from the inaugural membership list, and Anthropic on neither list as of 27 July 2026; implicator.ai places all four inside the older Coalition for Secure AI. That shape sets the reading, I think: the open agent-security stack is being assembled at the infrastructure layer, without the labs whose harnesses it aims to inspect.
Impact on your team
Two groups should care: teams choosing an agent framework, and teams whose prompts live in template files. If you adopt NOOA, decide now who reviews docstring diffs and label them prompt changes in the review checklist; your conventions treat a docstring edit as a comment and wave it through. I would ignore the membership roster and wait for independent evidence about the API, since every claim about the programming model comes from NVIDIA’s own paper.