Gradient Flow

If you check Ethics.Dev regularly, you may have noticed how often AI agents and cybersecurity have been colliding lately. Some of the stories sound straight out of movies: agents escaping test environments, communicating through channels nobody intended, finding vulnerabilities, stealing credentials, and reaching production systems. The part I think deserves more attention is less dramatic. AI is changing the economics of an attack. A human attacker has limited time and attention.

Subscribe • Previous Issues If you check Ethics.Dev regularly, you may have noticed how often AI agents and cybersecurity have been colliding lately. Some of the stories sound straight out of movies: agents escaping test environments, communicating through channels nobody intended, finding vulnerabilities, stealing credentials, and reaching production systems. The part I think deserves more attention is less dramatic. AI is changing the economics of an attack.

Last year that robotics had a data problem language models were lucky enough to avoid. Text, images, and code had accumulated for decades before anyone decided to train models on them. Robots had no comparable . That distinction is starting to look less clean. Recent systems have trained on enormous collections of human video. One company reported using more than a million hours. Another turned roughly 1,900 hours of first-person human footage into more than 18,000 hours of robot-format training data.

Subscribe • Previous Issues Last year that robotics had a data problem language models were lucky enough to avoid. Text, images, and code had accumulated for decades before anyone decided to train models on them. Robots had no comparable . That distinction is starting to look less clean. Recent systems have trained on enormous collections of human video. One company reported using more than a million hours.

In my conversations with developers and AI teams, I’m struck by how many are exploring moving more of their inference workloads to open models. I’ve touched on some of their reasons before , but here I want to look further ahead. I’ll admit my bias toward open source. I was around during the dot-com era and watched Linux displace Solaris, and open-source databases challenge commercial products like Oracle. I suspect we’re going to see something similar with AI models.

LLMs are built to generate text, but a surprising amount of AI automation does not need text at all. It needs a small decision: Is this spam? Which team should get this ticket? How urgent is it? We can force an LLM to return structured output, but it is still generating that answer token by token, which adds latency and cost when you do it millions of times. The model can usually handle these tasks, but I increasingly think we are using more model than the job requires.

Imagine a procurement agent checking inventory and deciding that stock has fallen below the reorder threshold. It places another order. The problem is that a large delivery was recorded a few minutes earlier, and the copy of the data the agent queried has not caught up. A dashboard running on stale data might show you the wrong inventory number. An agent running on stale data can place the wrong order.

Subscribe • Previous Issues Imagine a procurement agent checking inventory and deciding that stock has fallen below the reorder threshold. It places another order. The problem is that a large delivery was recorded a few minutes earlier, and the copy of the data the agent queried has not caught up. A dashboard running on stale data might show you the wrong inventory number. An agent running on stale data can place the wrong order.

Google recently agreed to pay $10 million for the internal business data of Spirit Airlines. The airline is bankrupt, but its emails, Teams messages, software, spreadsheets, and operating records apparently still have value. Google plans to use the material for product development and AI. Spirit’s flight attendants objected to the sale and sought additional protections for employee data, and a bankruptcy judge has delayed approval while the dispute gets sorted out.

Subscribe • Previous Issues Google recently agreed to pay $10 million for the internal business data of Spirit Airlines. The airline is bankrupt, but its emails, Teams messages, software, spreadsheets, and operating records apparently still have value. Google plans to use the material for product development and AI. Spirit’s flight attendants objected to the sale and sought additional protections for employee data, and a bankruptcy judge has delayed approval while the dispute gets sorted out.