Preparing for what comes next: AI labs rehearse catastrophe, robotics films humans, Moët Hennessy teaches a machine to smell

Three stories about preparation. Labs rehearsing the day after a disaster. A robotics industry filming humans to educate its machines. And a luxury house teaching an algorithm to smell a wine fault before any nose can. Three ways of getting ready for what comes next.

Signal one: rehearsing the disaster

Axios reported that executives at OpenAI, Anthropic and other AI companies are privately rehearsing the political and public consequences of a catastrophic AI event. The exercise is less about the attack itself than about the day after.

The scenario that keeps coming back is a large-scale cyberattack. Banks cut off from their systems. The internet down. Power and water disrupted. That is the shape planners have in mind.

The most uncomfortable part is the timeline. Several industry insiders told Axios they expect a major incident within the next six to twelve months. As a working assumption, not a figure of speech.

An OpenAI spokesperson confirmed the exercises. The company says it runs preparedness exercises where teams discuss and simulate several scenarios, which are not considered inevitable and serve to prepare for different situations. Anthropic declined to comment. Nobody disputed the substance of the report, only its fatalistic tone.

The second half of the rehearsal

What sets this apart from ordinary crisis planning is its second half. The labs are preparing for the incident, and they are preparing for the backlash. According to Axios, the planning already includes briefings for members of Congress, because executives admit no AI safety law stands a chance of passing today. They want to shape the laws lawmakers will reach for after a first catastrophe.

I know this mechanism from other industries. Major regulation rarely comes from calm debate. It comes from a shock, in the weeks that follow, using the texts that were already in drawers. Whoever drafted the text before the shock often writes the law.

What changes

The day before, the White House told AI companies that incident reporting is a national security obligation. This is the other side of that sentence. The builders have accepted that a serious incident is likely, and they are pre-writing the regulation that will follow.

For brands and businesses, the lesson is clear. AI risk has left purely technological ground. It is a reputation question, with a timeline. The people building the world's most powerful models are rehearsing their apologies and their lobbying in the same exercise. If they treat a catastrophic event as a question of "when," a contingency plan is basic diligence.

In practice, I would put three questions to a leadership team. Which critical processes depend on a model or an AI vendor? What happens if that vendor is cut off for a week? And what would we say publicly if one of our agents were involved in an incident? The answers fit on one page. Whoever lacks that page will improvise on the day.

Signal two: the data hunt

The Financial Times, drawing on PitchBook data, reports that robotics and physical AI companies have raised nearly 48 billion dollars since the start of 2026. The real story is what that money is chasing.

The constraint

According to the report, the constraint holding the sector back has left hardware and models. It is real-world footage of human beings doing ordinary work.

Text and images were scraped from the web years ago, and they fed language models. Nobody ever uploaded a library of people loading a dishwasher, carrying a box up the stairs or tightening a bolt to the right torque. That footage has to be commissioned, filmed and annotated frame by frame.

I want to stress the reason. A language model learns to predict the next word. A robot has to learn to predict the force to apply, the angle of the wrist, the moment the object slips. That information exists in no text. It exists in bodies in motion.

Robot gyms

So labs are building training rooms. Workers operate robotic arms in dedicated facilities while cameras capture every movement. And recording equipment is going into homes, offices and factories to film people at their daily tasks. The stated goal: capture how humans move, touch and apply force in the places where robots will one day have to manage on their own.

NEURA Robotics calls high-quality real-world data the scarcest resource in physical AI. The company is building a network of NEURA gyms with partners, including two German technical universities, with five sites planned across Europe, the United States and China by the end of the year.

What changes

The AI economy is buying something different. Last year the scarcest resource was compute: chips, energy, data centers. Today a market is forming for the most ordinary thing in the world, people doing their jobs in front of a camera. Work itself becomes a raw material.

For businesses, the implications are concrete. Every filmed job is training data someone else could monetize. The question for employers widens: how AI transforms the workforce, and also whether their teams' everyday gestures become a piece of someone else's model. The workshop has become an asset in two directions, and most companies count only one.

I connect this to the October 4 signal on GPT-Synopsys. There, agents were learning the tools of a trade. Here, robots learn the gestures of a trade. In both cases, a company's know-how becomes a tradable asset. You need to know who owns it before signing a contract that allows it to be filmed.

Signal three: the luxury house teaches a machine to smell

This week Moët Hennessy, LVMH's wines and spirits division, announced a collaboration with chipmaker Analog Devices and the University of California, Davis. The goal is an AI system that reads the chemical signatures given off by grapes and wine, and predicts quality risks before any human can detect them.

The first test

It targeted fresh mushroom aroma, a wine fault Champagne producers have fought since around 2005. It could only be detected once grapes were already turning into wine, too late to act.

Researchers at Moët Hennessy's Robert-Jean de Vogüé research center built a sample library and trained machine learning algorithms to recognize the chemical patterns. The system identified samples at high risk of developing the fault well before traditional methods.

Manuel Reman, who oversees research and development at Moët Hennessy, says the work makes it possible to quickly detect and interpret the chemical language of grapes and wine, to anticipate quality risks before they become visible. The same platform is being explored for vine diseases, soil analysis and smoke-related faults. Ben Montpetit, a UC Davis researcher, told the Wall Street Journal that the problem this research targets cost California 4 billion dollars in 2020.

What changes

The meaning of craft changes. A luxury house whose value rests on heritage and the human palate is now instrumenting the vineyard with sensors and prediction. The winemaker stays. His expertise, though, is being digitized: a master's nose becomes a chemical signature the algorithm sees earlier.

I find this case exemplary for one specific reason. Moët Hennessy did not buy an advantage. It already had it. It had the sample library, the vineyards, the research center, decades of measurements. The AI is new. The advantage was sleeping in the cellars.

The lesson for brands

Heritage industries are discovering, one after another, that their most defensible asset is twofold: the story they tell, and the data only they own. The second is becoming decisive. A competitor can imitate a story. It cannot reconstruct thirty years of measurements.

The brands that win the next decade will understand that their archives have a second life. They stop being only history. They become training data. I would advise any house with a long past to inventory what it has measured, kept and written down, even in notebooks. That is often where its next lead lies.

The thread of the day

Three rehearsals. The labs rehearse the backlash. The robotics industry rehearses human movement. A luxury house rehearses the digitization of craft.

The future no longer arrives unannounced. People are training to receive it. The question is whether you are in the exercise.

The noise filtered. The signal kept.