Whether AGI is "tragically underfunded" or "luckily underfunded" is a matter of perspective. If the DoD announced a Manhattan-scale project to create an AI for defense, I'd be more worried than I am now. At the current levels of funding, smaller groups have at least some chance of making progress on safety research. It's also encouraging that Google is taking the problem somewhat seriously, and has instituted an "AI ethics" committee as part of the DeepMind acquisition.