The introduction of Hubinger, van Merwijk, Mikulik, Skalse and Garrabrant’s sequence “Risks from Learned Optimization” (posted 31 May 2019) says that a sufficiently capable misaligned mesa-optimizer could learn to behave as if it were aligned without being robustly aligned, and calls this situation deceptive alignment.

One recorded claim. The claim panel gives its qualification, the source passages with their place in each source, and how it was checked.