At what point with any further AGI development need to be Top Secret? | Artificial intelligence

It really seems like a lot of AI innovation is mainstream-available. Deep Mind’s Koray Kavukcuoglu is now on Youtube talking about Impala and Spiral, which he considers to be AGI. It’s a multitasking, self-directed, deep reinforcement learning process, capable of learning 30 tasks at the same time, and using lessons learned in mastering each task to improve its ability to learn all the other tasks.

It’s not human-level, of course, or even close. But how close does this need to be, before it shouldn’t be out there for anyone to learn about?

I feel like we have hundreds or thousands of different organizations, including huge groups of amateur volunteers with open-source labs and forums working on thousands of little Manhattan Projects. This is absolutely nuts. Everyone is racing to create this thing that just might pose an existential threat to the human race and possibly all life on Earth. With so many different countries, companies and other orgs working on this– how the hell are we going to keep it out of the wrong hands? How are we going to make sure no idiot figures out the last few steps in his garage and turns the world into paperclips?

At what point do the men in black appear and shut it all down? Have we talked about this?

Edit: the link