Security BSides London 2018

Delayed execution is a concept of significant interest to attackers, who seek to use it so that their malware is able to bypass the analysis period of sandboxes and antivirus emulators. Historically, techniques used to delay execution have included Windows API calls, and short, simple loops involving assembly, counters, or loading libraries. However, security tools are increasingly able to detect and prevent these techniques, using methods such as accelerating time, returning false tick counts, intercepting API calls, and performing multipath execution. As a result, attackers are constantly striving to find new and creative ways to delay execution. Delayed execution is also of some interest to defenders, who try to implement it, in either manual or automated solutions, in order to frustrate the attack models of bots, botnets, and spammers.

Enter the timelock puzzle - a relatively unknown cryptographic construct whereby a puzzle is presented, the solution to which requires a certain amount of time or computational effort. Historically, timelock puzzles were proposed for benign applications, such as sealed auction bids, escrow, and the timed release of confidential information. However, they provide an interesting method of delayed execution which to date has been underexplored in security research, particularly as an offensive methodology. Specifically, they may present a significant challenge in malware detection and analysis, particularly for automated solutions such as sandboxes.

In this talk, I cover the history of timelock puzzles and their proposed applications for offence and defence, and examine some case studies. I then demonstrate several timelock puzzles which I have developed, including some novel constructions, and show through demonstrations how they can be weaponised - including both process hollowing within executables, and within VBA macros. For each construction, I explore the advantages and disadvantages for both attackers and defenders, and explain how they work, and why. I then turn to prevention and detection, presenting a heuristic model for generic detection of timelock puzzles, and cover the defender's perspective in the form of attacks against timelock puzzles, including parallelisation, predictability, and enhanced computational processing.

I then cover the challenges and feasibility of using timelock puzzles for good, discussing some of the models presented in previous work and a real-world case study where timelock puzzles could have been used to significant effect, break down a proof-of-concept defensive timelock puzzle I created, and some of the issues identified with it from an attacker's perspective.

Finally, I assess the practicality of timelock puzzles for both attack and defence, share some lessons learned from this research, and outline suggestions for future research in this area.