Frame Challenge: this is not efficient
Sending drones straight from distributer to household with a single package is the very definition of an inefficient system. It takes an \$18,000 drone an entire round trip to deliver a single 10lb package. In contrast, a fully loaded \$30,000 delivery truck can carry 800 10lb packages out to a neighborhood, and then drive just a few feet from door to door dropping them off instead of making a return trip for each package. This means that a single truck can do the work of hundreds of drones with a similar capacity.
Trucks also have more range than a drone. A commercial grade drone carrying 10lb can expect about 20 minutes of flight time and a top speed of about 40-60mph. That is a theoretical maximum range of 7-10 miles; though, to account for battery degradation, potential motor failures, weather, etc. the actual safe operational distance of a drone is more like 5 miles whereas trucks can travel for hundreds of miles between refueling.
Drones break down more easily than trucks. The maintenance requirements on drones per package is also much higher. If we assume you need 200-400 drones to replace 1 truck, then that is 800-1600 times as many motors. Even if drone parts are individually smaller, cheaper, and simpler to work on than the truck motors, the overall maintenance cost and labor will still be orders of magnitude higher. In terms of human labor, a single truck driver and the occasional mechanic is much cheaper than the maintenance crew it takes to keep hundreds of drones going.
Then there is the fuel issue. Flying is less fuel efficient than driving, BUT if your drone base is close enough to consumers, they can save on efficiency by not carrying 799 extra packages to more than 1 individual address. In our world, drones can save delivery companies a lot of money on "last mile delivery" because delivery trucks waste so much fuel driving from 1 place to the next carrying all those extra boxes, but in your world, where every house gets a package every time, you eliminate most of the wasted fuel you see in producer-to-consumer shipping. This means trucks suddenly become more fuel efficient than drones because there is not a lot of driving around to spaced out homes to unload the whole cargo load.
The other advantage of drones in modern distribution is time. Again, with producer-to-consumer shipping, drones save a lot of time when it comes to loading a truck and driving it around all day until your package finally reaches you, but this is a non-issue for your world. If you are sending the same package to the same place on a regular schedule, then having a fast response time to delivery is not important. It could take a long time to get a package from the factory to someone's home, and not be an issue as long as packages are sent at a regular interval, they will arrive on a regular interval.
So, your delivery system would be hand-over-fist more efficient with trucks than drones.
What about a robotic delivery truck service?
Okay, so let's say we are still using trucks, but they are operated completely by robots instead of people. You have now solved the delivery system, but NOT the interception issue. The big problem with delivery is that once you drop a package off at someone's doorstep, you are leaving it in the open for anyone to steal. In the parts of the world where hunger is the biggest issue, the problem is not a lack of food, but warlords/cartels/etc. intercepting shipments.
Robot delivery systems are super easy to rob. A guy could literally just drive around behind a delivery truck (or if you do go with drones anyway, learn the delivery schedules) and pick up packages as they are dropped off. People won't even try to grab their own food allotments because they know if they do, some guy with a manchette is going to show up in a few minutes to cut your head off for "stealing from the cartel". As far as the AI is concerned, it's delivered packages to every home. Mission complete, hunger is solved! ... only no one actually got their food because you are trying to feed some lawless s***-hole part of the world where criminals just see one more easy opportunity for exploitation.
Starvation will still be common in such places because if they want food, they have to buy it from the local warlord/cartel for exorbitant amounts of money that they probably can't afford because the warlords are also reselling the food as extra rations in richer communities, keeping parts of it for themselves, and using it to feed their off-grid human trafficking/drug operations where they can not report where people live; so, only small portions of the food intended for the area actually makes it to those people.
The other common cause of starvation is warzones. Places where shipping gets blockaded to prevent the mobility of enemy assets. In a warzone, soldiers are not going to waste time trying to figure out if that is a food drone or a an enemy attack drone... and if they do decide to ignore food drones, it just encourages the enemy to disguise their attack drones as food drones. No, a no-fly-zone is a no-fly-zone, AI or not. So, food distribution will still be cut off just like it is now.
If you want an AI to solve world hunger, they don't need to improve the distribution system. Humans are already really good at this. Instead, you need to improve law and order so that your distribution systems actually work.