For 6G, the ambition is broader. AI may support closed-loop decisions on beams, scheduling assistance, mobility, power and energy saving, service assurance, and resource orchestration. But “AI-native” should not imply that every control decision is handed to an opaque model. The standards challenge is to define the frameworks that allow AI-assisted functions to be configured, validated, monitored, coordinated, and safely reverted, while leaving model design and implementation to vendors and operators. That matters because 6G is expected to operate in a much more complex environment, with denser deployments, higher and more directional spectrum, integrated sensing, terrestrial and non-terrestrial coverage, distributed compute, and strict energy-efficiency goals.

Operational validation of a 6G network cannot be treated as a one-time architectural sign-off. A 5G network is already automated, but much of its behavior still follows predefined policies, engineered parameters, and well-understood control loops. A 6G network is expected to go further, with AI/ML-assisted functions adapting to live changes in radio conditions, traffic, faults, energy constraints, and service objectives. That does not mean the network acts without limits or “changes its own mind” in an uncontrolled way. It means some decisions may be made inside controlled closed loops, within operator policy, safety limits, fallback procedures, and intent-defined boundaries. The architect defines those guardrails, but the operations team must continuously validate whether the live network is behaving correctly. At 2 a.m., if the system routes around a fault, changes beam behavior, or reallocates resources, operations needs to know whether that was a valid adaptation or an overconfident decision. The validation challenge is no longer just, “Was it configured correctly?” It is, “Is this adaptive system still acting safely, explainably, and in line with operator intent?”

I don't have a clean answer to that, and I'd be suspicious of anyone who claims they do this early. But those are the real work items hiding under "AI-native," and they want answering before the standard freezes, not after.

Testing early is the key to meeting the timeline 

This is the part of the schedule with no formal milestone next to it. When the windows are this compressed and the system this complex, the challenge won't be writing the specifications. It'll be proving the designs work before anyone pours silicon and capital into them.

Look at what has to be validated before any 6G deployment. Peak data rates moving far beyond 5G. AI workloads becoming part of normal network operation rather than an occasional optimization tool. Radio behavior in higher and more directional bands where propagation is harder to model, more sensitive to blockage, and less forgiving of bad assumptions. Energy efficiency becoming a design condition, not a nice-to-have, because a network carrying far more traffic across terrestrial, aerial, and non-terrestrial nodes will not be economically or environmentally viable if it scales power consumption at the same rate as capacity. Every one of those behaviors has to be emulated under conditions that look like the real network, because the alternative is discovering the problem in the field, and the field is the most expensive place to find it.

This is where I should admit my bias, the whole point of design, emulation and test is to close the gap between what a system looks like on paper and how it behaves once it's real, and to close it early enough that the fix is cheap. That's the work we do, so weigh it accordingly. On an ordinary program you might be able to leave validation to the back end. Not this one. With concurrency built into the standards process itself, validation has to run alongside design from the start. The groups that come out of this with real standards influence and a real shot at time-to-market won't be the ones who tested late. They'll be the ones who were already prototyping and continuously stress-testing candidate designs while the ink was still wet, then feeding evidence back into the next round.

Yes, that reads like a pitch, so let me strip it down. The ITU deadline is fixed. The 3GPP freeze is fixed. The engineering windows are six quarters apiece. Migration is undecided. The network is AI-native, which makes it harder to trust than anything we run today. You can't buy more time. The one lever you actually control is how early and how hard you innovate, test and validate.

The clock is real now in a way it wasn't twelve months ago. The operators and vendors who treat testing as where you start, rather than where you end up, are the ones who'll still be on schedule when 2029 arrives.