There's a fine line between having enough power to do something useful and having all the power to do anything you want. Java provides the foundation for a secure environment in which untrusted code can be quarantined, managed, and safely executed. However, unless you are content with keeping that code in a little black box and running it just for its own benefit, you will have to grant it access to at least some system resources so that it can be useful. Every kind of access carries with it certain risks and benefits. The advantages of granting an untrusted applet access to your windowing system, for example, are that it can display information and let you interact in a useful way. The associated risks are that the applet may instead display something worthless, annoying, or offensive. Since most people can accept that level of risk, graphical applets and the World Wide Web in general are possible.
At one extreme, the simple act of running an application gives it a resource, computation time, that it may put to good use or burn frivolously. It's difficult to prevent an untrusted application from wasting your time, or even attempting a "denial of service" attack. At the other extreme, a powerful, trusted application may justifiably deserve access to all sorts of system resources (e.g., the filesystem, process creation, network interfaces); a malicious application could wreak havoc with these resources. The message here is that important and sometimes complex security issues have to be addressed.
In some situations, it may be acceptable to simply ask the user to "OK" requests. Sun's HotJava Web browser can pop up a dialog box and ask the user's permission for an applet to access an otherwise restricted file. However, we can put only so much burden on our users. An experienced person will quickly grow tired of answering questions; an inexperienced user may not even be able to answer the questions. Is it okay for me to grant an applet access to something if I don't understand what that is?
Making decisions about what is dangerous and what is not can be difficult. Even ostensibly harmless access, like displaying a window can become a threat when paired with the ability for an untrusted application to communicate off of your host. The Java SecurityManager provides an option to flag windows created by an untrusted application with a special, recognizable border to prevent it from impersonating another application and perhaps tricking you into revealing your password or your secret recipe collection. There is also a grey area, in which an application can do devious things that aren't quite destructive. An applet that can mail a bug report can also mail-bomb your boss. The Java language provides the tools to implement whatever security policies you want. However, what these policies will be ultimately depends on who you are, what you are doing, and where you are doing it.
To fully exploit the power of Java, we need to have some basis on which to make reasonable decisions about the level of trust an application should have. Web browsers such as HotJava start by defining a few rules and some coarse levels of security that restrict where applets may come from and what system resources they may access. These rules are sufficient to keep the waving Duke applet from clutching your password file, but they aren't sufficient for applications you'd like to trust with sensitive information. What if you want to implement a secure applet to carry a credit card number to the mall, or more likely the credit-card company? How are people to trust that the applet they are using is really secure? If it's named the "Bank of Boofa" applet, how do they know it's legit?
You might think of trusting only certain hosts for these kinds of applications. However, as Java class files begin to fill the Net, the situation will become more complicated. Hosts can be impersonated. If your communications pass through an untrusted network, you can't be sure you're talking to the intended entity. Furthermore, class files may need to be cached or retrieved through complicated distribution mechanisms. For these kinds of applications, what we really need is a mechanism for verifying the authorship and authenticity of an item and making sure that it has not been tampered with by the time that you received it. Fortunately, this is a problem solved a while ago by your friendly neighborhood cryptographers.
Digital signatures provide a means of authenticating documents. Like their inky analogs, they associate a name with an item in a way that is supposed to be difficult to forge. Unlike pen on paper, though, electronic digital signatures are actually difficult to forge when used properly. By their nature, digital signatures also provide the benefit that, if authenticated, a document is known not to have been altered in transit. In other words, you can't clip out a digital signature and attach it to a new document.
The details of cryptography are a bit beyond the scope of this book but the basics are important and interesting. Digital signatures are one side of the coin of public-key cryptography. Public-key algorithms rely on the fundamental mathematical difficulty of factoring arbitrarily large numbers. In a public-key system, there are two pieces of information: a public key (as you might have guessed) and a private one. These keys have a special asymmetric relationship such that a message encrypted with one key can be decrypted only by knowing the other. This means that by giving you my public key, you can send me messages that only I can read. No one else, including you, has enough information to decrypt the encoded message, so it's safe to send it over untrusted networks. Now, by reversing this process, I can encrypt something with my private key so that anyone can use my public key to read the message. The important thing in this case is that the task of creating such a message without the private key is just as difficult as decoding the message in the first scenario. Since no one else knows my private key, only the real me could have sent the message. This is the basis for digital signatures. For Java, this means that we can tell a browser "I trust applets signed by John Doe"; if the browser succeeds in decoding an applet using John Doe's public key, it knows that the applet really came from John Doe, and therefore can be allowed greater privileges.
 See Bruce Schneier's encyclopedic Applied Cryptography (John Wiley & Sons).
This process can be used to authenticate Java class files and other types of objects sent over the network. The author of a class signs the code with a digital signature, and we authenticate it when we retrieve it. Now we know that we have the authentic class, or do we? There is one problem that a digital signature alone doesn't solve: at some point we still have to assume we have the author's authentic public key. This is where a key-certification agency comes into play.
A key-certification agency validates a key by issuing a certificate that lists a name and an official public key. The certificate is signed with the agency's own digital signature. The agency presumably has a well-known public key to verify the certificate. Of course, this doesn't solve the problem entirely, but it reduces the number of people you have to trust and the amount of information you have to transport reliably. Presumably the agency is a reputable organization, its private keys are well guarded, and it certifies keys only after some kind of real-world validation such as person-to-person contact.
The most recent Java release (1.1) contains the tools you need to work with signed classes. You can sign Java classes; you can tell the HotJava browser whose classes you trust (and how much you trust them). Other browsers, like Netscape Navigator, should support signed classes in the future. You can also use the security API in your own Java programs to handle sensitive data safely. The important thing is, as always, to know who you are dealing with and what kind of software and security you have in place before sending any kind of confidential information over the Net. Don't become paranoid, just keep yourself informed so that you can weigh the risks and the benefits.