I haven't seen anywhere in the spec that required a security manager to be installed (at least, not yet... still reading). I have seen containers that optionally allow you to configure a security manager. I'd bet that the spec is just trying to make it clear what a bean provider can depend upon. If a bean provider used reflection to get access to stuff he/she shouldn't, the bean may not be portable (or the deployer/sys admin might not be able to do their job). The interfaces remote/local declare what you are allowed to have access to, using reflection on anything except what you are allowed to have access to isn't wise (in EJB, or in
applets, or in any other Java code where security restrictions an apply).