# How to use an Aviation Safety Management System for AI

- **Series:** Aviation for AI, Part 2
- **Author:** Samuel Chandra (Founder & CEO, Deepsky)
- **Canonical:** [https://www.deepskyai.com/newsletter/how-to-use-an-aviation-safety-management-system-for-ai](https://www.deepskyai.com/newsletter/how-to-use-an-aviation-safety-management-system-for-ai)
- **Subscribe directly:** [https://www.deepskyai.com/newsletter#updates](https://www.deepskyai.com/newsletter#updates)

After my realisation in my first post that actually - Aviation for AI is probably a much more consequential pursuit than AI for Aviation, I thought it would be useful to flesh out what that could mean practically. Firstly though, it’s worth answering - why would you need a safety management system in AI? Well is the same reason you need one in aviation:

Adopting a safety management system allows you to do much more risky things, without the downside. Therefore, robust safety management allows you to capture far more upside for your business/activity/organisation.

Aviation took a while. It was always risky and people died left right and centre, when new technologies were invented, it usually opened up a whole lot more possibilities (like jet engines or  pressurisation) and so passenger numbers increase and loss of life too - but every time that happened, safety systems were improved and downside was capped.

In enterprise, people really want to realise the promise of AI - all that alpha they can capture by doing otherwise impossible things with agents and intelligent units just running around their company. But they’re also scared - scared that it will release sensitive information, shut down a critical system or wipe a database. And they should be scared. Rolling out this kind of technology - with its ability to reason and act autonomously with the same cyber and resilience thinking you had five years ago will result in disaster. But, if you realise that we are dealing with something fundamentally different (to old technologies) but not too dissimilar to these other autonomous, reasoning machines we are familiar with (known as humans) then aviation supplies the prior art of making this all extremely safe.

Maybe OpenAI, Anthropic and Meta could have taken a few minutes to think through this prior to reaching for the levels of risk they have already reached for.