The sky can scale faster than its safety system.

What I see

In ICAO’s UTM Framework, Edition 4, I see an ecosystem built around collaboration between operators, service providers and authorities, supported by automated information exchange. Risk assessment and contingency planning sit within that picture. The aircraft is only one part of the system. [ICAO]

EASA’s AI Roadmap 2.0 places a human-centric approach at the heart of aviation AI. I read these documents together as a challenge: connect more activity without losing the ability to understand, challenge and control it. That connection is my interpretation, not a joint regulatory conclusion. [EASA]

My assessment

A successful demonstration answers a bounded question: could this operation work under those conditions? I would not let that answer silently become proof that a larger network will remain resilient when weather changes, a supplier fails or several operators need the same limited airspace.

For me, the value of In-Time Aviation Safety Management Systems lies in the decision chain: notice the change, understand the risk, identify who can act, and check whether the intervention helped. AI and machine learning may strengthen that chain through pattern recognition and forecasting. More predictions alone do not complete it.

I would ask an operator to show how a warning becomes a decision. Who receives it? How much useful time remains? Who has authority to restrict activity? What happens when the model and the operational team disagree? These are my assurance questions, not additional regulatory requirements.

What concerns me

I am concerned about confidence travelling faster than evidence. A persuasive dashboard can conceal stale data, incomplete coverage or several services relying on the same fragile input. Adding automation can make a dependency less visible without making it less consequential.

Through my Adversarial Lens, I ask: if a trusted input were manipulated, would the system recognise the contradiction—or spread it? This is a defensive scenario, not an allegation against any organisation. The safety issue is the consequence of misplaced trust, whether caused by malicious activity, an ordinary fault or poor design.

My concern does not establish that a particular LAE programme is unsafe. These sources cannot support a worldwide judgment about readiness, commercial viability or deployment of complete IASMS capabilities.

What I would watch next

I would look for published evidence of operations coping with degraded data and communications, clear intervention authority across organisations, and measures of whether alerts arrive early enough to change the outcome. I would also look for model monitoring after deployment and records showing that lessons lead to changed controls.

My assessment would become more positive where repeatable, independently evaluated results demonstrate those capabilities under representative conditions and within a clear approval scope. Announcements and target dates would not carry the same weight.

My practical starting point: select one important risk and trace it from signal to accountable action to evidence that the control worked. If that chain is unclear, I would prioritise strengthening it before treating greater scale as proof of progress.