The New Spam: Machine learning can be used to write human-like tweets

Phishing messages typically get 5-10% response rates, but a new system has boosted its rate to 40%. John Seymour and Phil Tully, two data scientists from the security company ZeroFOX, presented their system SNAP_R at Black Hat, a Las Vegas conference on cyber-security, on August 4. SNAP_R uses a deep neural net to study a person's past tweets and then mimics that person's writing style using a Markov model, generating a phishing tweet. So far, there is no reason to think that criminals are using a similar system, but Seymour and Tully's work show how it might be done. You can read more in this article from MIT Technology Review or Seymour and Tully's paper.