Top 10 Unexpected AI Systems Hacks that we did not see coming!

No ratings

Presented at CactusCon 13 by

Gopi Ramamoorthy

Aruneesh Salhotra

In the early 2000s, when a new wave of internet search engines hit the market, there were a flurry of search engine hacks surfaced. The major ones are based on Google search engine features and they were called Google dorks. These dorks exposed vulnerabilities, sensitive data files and config data files to the public. With AI rapidly getting deployed and adopted, there is a trend of AI hacks, an equivalent of Google dorks surfacing in the wild since last year. A subset of these hacks are called adversarial attacks. There have also been unexpected ways of exploiting AI based systems. The speakers will talk on different types of adversarial attacks, trends and the threats posed by either insiders or external bad actors. In the last two years there have been 50+ different AI adversarial hacking methods that have surfaced in the industry. The presentation will focus on high impact unexpected methods of hacks in multiple sectors including government, finance, healthcare, transportation and education. The first half of the presentation will discuss wide spectrum of attacks from non technical hacks to highly technical attacks. Interestingly, the speakers will share the analysis of real incidents that occurred in the last three years. The second half of the presentation will be on what went wrong, preventive techniques, and evolving platforms. The summary of lessons learnt and short term and long term recommended actions will be presented in the final part of the presentation.