I want to pretend to control target@example.com and the legitimate owner of this address is an OpenPGP user who has published a keyring on the public keyservers.

1) I create a keyring and add a single uid with my real name and target@example.com

2) I download the public keyring for the legitimate user target@example.com and extract the encryption subkey.

3) Even though I don't know the private key I can add this public key as the encryption subkey to the keyring created in step #1.

4) I publish this keyring on the public keyservers so that you will find it by querying the fingerprint I give you when we meet.

5) You send email to the real user target@example.com which they are able to decrypt and respond to. Of course there could be some confusion since the real user is not expecting an email which presumably talks about verifying keys.

6) Since the mail was decrypted and responded to, you sign the key and return it to me.

7) I revoke the certification on the encryption subkey I borrowed from the real user and add a new encryption key which I create.

8) People who trust your signature encrypt mail to target@example.com with the false key I've published.

> What kind of attacks is this practice vulnerable to?

So long as the trust only translates to the very limited use cases then there is no vulnerability. Those limits basically mean that no trust should be assumed by anyone that did not participate in the WoT in question. This particular style of WoT means anyone wanting to retain anonymity needs to skip it, or am i missing something?

Mainly I think all WoT models i've seen thus far are too susceptible to sybil attacks (impersonation) and as a result instill bad habits.

you only need to look at people use of pgp today to see how much WoT's would be a failure if used. pgp.mit.edu still aint got ssl. journalists are linking to nonssl links for their key via twitter t.to urls