On 7 September Google and Cathay Pacific said they were taking an AI contrail-avoidance trial from a first operational batch into a larger second phase on Asian and transpacific ultra-long-haul sectors, with Contrails.org joining as a science partner. The first run, begun in late 2025, targeted more than 100 flights. More than 80 followed avoidance routes. Google’s own satellite analysis estimated those flights cut the warming impact of the contrails they would otherwise have made by about 40 percent. Hong Kong–Singapore did more than half the work. Predictions reached the deck through Cathay’s electronic flight folder over cabin Wi-Fi. The companies present persistent contrails as a large slice of aviation’s total climate effect. Google’s Kemal Armada used a figure of about one-third to present the fix as a small altitude change, the same family of tweak a crew already makes for turbulence. That is the claim that has to survive a dispatcher’s night.
It is a forecast of a thin, cold, humid layer, a request to fly two thousand feet off the planned cruise, and a judgement about whether the extra fuel burns more carbon than the ice cloud would have warmed. Forecast error is the first risk. A layer that was going to persist for hours may be a patch that dies in sunlight. A layer that was going to be harmless may sit exactly where the next sector is already full. Additional fuel is the second. A long-haul 777 or A350 that steps up or down for an hour is burning kerosene to buy a non-CO2 credit that exists only in a model. If the model is wrong, the airline has spent carbon to miss a cloud. ATC is the third. Asia-Pacific tracks and the Pacific organises traffic by flight level. A request to leave the assigned band is a conversation with a controller who is protecting separation, not a climate spreadsheet. Who decides is the fourth. Dispatch can file the avoidance. The commander can accept or refuse. A saturated corridor can refuse for them. Europe has already started metering the same effect. Cathay is trying it where the sectors last sixteen hours and the weather file is stale before top of descent.
The honest product of phase two is not another 40 percent headline. It is a published fuel penalty per avoided contrail-hour, a hit rate for the forecast against the satellite after landing and a note on how often ATC said no. Until those three numbers exist beside the trial, the industry has a useful experiment and a temptation to treat a dispatcher’s altitude request as a substitute for sustainable fuel. The cloud is real. So is the burn that avoids it.