Human Factors in Security Phishing, Scam, Leaked Credentials Phishing and spoofing Why email spoofing is still possible? How could spoofing/phishing emails penetrate the current defense? How to build robust and usable phishing detection and alert systems Measurement + User study Poor defense: SPF/DKIM/DMARC, 4%~48% adoption rate Spoofing indicators are largely missing Mobile phishing is even more serious Gmail
Human Factors in Security (Cont.) Phishing, Scam, Leaked Credentials Understanding the persistent threat from leaked credentials How often do users reuse or modify passwords across services? How quickly do users change the reused (leaked) passwords? Are modified passwords predicable? Empirical approach instead of user studies Data-driven: 28M users, 62M passwords, 107 services 52% of users reused or modified passwords across services Email and shopping passwords mostly reused Guessing algorithm: 16 Million passwords cracked in < 10 guesses
Adversarial Machine Learning in the physical domain Stop Sign Fool a machine learning system is relatively easy Most work focuses on the “digital domain” Machine learning cannot (does not) reason like human New challenges in the physical domain Various sources of errors make the attack more difficult View angle, distance, quantization, resolution Attacks require training data from the physical world, expensive Can we model the differences between digital and physical world? How to defend against adversarial attacks? Yield Sign + =
Other Projects GPS spoofing to attack mobile navigation systems (self-driving car) In preparation Crowdsourcing social media data to detect security events CIKM 2017 Mobile deep links to hijack web-to-mobile communications USENIX Security 2017 Detecting collusion between mobile apps Asia CCS 2017, MoST 2017 The social aspects of mobile payments ICWMS 2017, GROUP 2018