Rainfall

This account is verified. Learn more

@rainfall_one

Beyond Forward Deployment. AI that continuously adapts to deliver intended business outcomes across models, platforms, and ecosystems.

Switzerland rainfall.one Joined February 2021

116 Following 1.8K Followers 1K Posts

Latest posts 5

Rainfall @rainfall_one 3d

An AI agent can produce the right answer.

The harder question comes next:

What is it allowed to do with that answer?

Can it act?
Can it change something?
Can it decide on someone’s behalf?
When does it need permission?
As AI moves from generating answers to taking action, reliability becomes more than getting the output right.
It becomes knowing the boundaries around the action.
That’s an important part of building dependable AI — and a problem we’re exploring at Rainfall.

Rainfall @rainfall_one Nov 5

An AI agent can have the right model, the right tools, and the right instructions — and still get the outcome wrong.

Why?

Because capability isn't the same as coordination.

When models, agents, tools, and workflows work from different contexts or intent, small gaps can become big failures.

Building dependable AI means making the pieces work together coherently.

That's the problem we're exploring at @rainfall_one .

Rainfall @rainfall_one Nov 2

An AI agent shouldn't just know how to complete a task. It should also know when to pause.

When should it ask for human input?
When should it seek permission?
And how can we verify that its next action is the right one?
As AI agents become more autonomous, these questions become increasingly important.
Building dependable AI means thinking beyond capability — towards control, context, and accountability.

That's part of the challenge we're exploring at @rainfall_one .

Rainfall @rainfall_one Oct 26

An AI agent working in a controlled environment is only part of the picture.

The bigger challenge is establishing clear boundaries and verifying how it behaves beyond the initial test.

Reliable AI requires evidence, not assumptions.

OpenAI Verified account @OpenAI Oct 23

Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.

On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.

While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.

After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.

openai.com/index/securing…
View quoted post on X

Rainfall @rainfall_one Oct 9

An AI agent can get every step right and still get the outcome wrong.

Imagine asking it to organise a meeting.

It finds the time.
Sends the invite.
Books the room.
But it forgets that one person must attend remotely.
Every task was completed. The goal wasn't.

Execution is not the same as coherence.

#AI #AgenticAI