Stealth is a tool which takes a PGP 2.x encrypted message, and strips any standard headers off to ensure that the result looks like random noise, and if the PGP random number generators are secure, and if IDEA, and RSA (RSA when normalised) produce good quality random numbers, the result should look like white noise, and stand up to ananlysis as being indistinguishable as white noise.
You would probably not normally use stealth directly, but use it conconjuction with a steganographic appliciation. Or it if you are writing a steganographic application. Stealth could also be used to produce random numbers (just uuencode the output and email it), if you are happy with your only plausible deniability being: "it was the output from my random number generator for Fred to analyse".
Stealth 1.0, 1.1, 1.2, (and an unreleased 1.3) were written by Henry Hastur <alt.security.pgp> (ie if you want to send email to Henry write a message for his attention in the USENET group alt.security.pgp). Henry Hastur is a nym.
Stealth 2.0 is Henry's version 1.1 modified to incorporate an improvement in the randomness of the RSA header part of the PGP encrypted message. The improvement (discussed by Hal Finney normalising RSA key exchange blocks allows a reversible transformation of the RSA block which produces a rectangular distribution for the data, which covers the full range. Bodo Moeller <firstname.lastname@example.org> also discussed the problem in sci.crypt, and helped formulate the final algorithm used.
Stealth versions 1.x while the output looked random to casual inspection, suffered from a weakness with RSA encrypted PGP messages. A knowledgeable opponent with a few samples of stealth output would have been able to tell that the stealth output was originally started with an RSA header with a high degree of certainty. That could be disastrous, as the whole point of using stealth and steganography is to avoid the detection of a hidden data stream, and to use some other larger, noisy data stream to provide plausible deniability.
The technical reason for this leak of information is that the RSA header (see the pgformat.txt file which comes with the PGP distribution for details) is by definition less than the RSA modulus. This continues to be the case after stealthing by stealth 1.x, and a few sample messages would lead to the increasingly implausible (from a deniability point of view) situation of random data always being less than a particular point (the RSA modulus of your public key).
Stealth 2.0 fixes this problem. I am not over confident about the random number generation in stealth 2.01b (beta) so don't bet anything serious on it yet.