Notice

Competition summary: predictive cyber analytics

Updated 6 September 2018

1. Background

Traditional cyber security methods only respond to known threats. However, as our understanding of adversaries and attack patterns improves, and increased computing power and data growth continues to drive the Artificial Intelligence revolution, new possibilities are emerging to get ahead of threats and predict future cyber attacks.

This DASA competition is interested in novel approaches to cyber security which can predict future threats to aid our defence strategies in preventing them.

For Phase 1 of this competition, DASA is looking for proof-of-concept technologies above Technology Readiness Level (TRL) 2.

Additional funding is anticipated to be available for future phases to further develop technologies to higher TRLs towards integration into a complete deployable system.

Computing infrastructure is a key component of nearly all modern defence systems and provides another attack surface for adversaries. Cyber security has been in an arms race for decades, with hackers continuously exposing new vulnerabilities and developers racing to patch them. Approaches to cyber defence have historically been reactive, relying on whitelists, known (virus/malware) signatures, or more recently on broader machine-learning detection methods. Such reactive methods are forensic or, at best, real-time. There has been limited effort in predicting events related to a cyber attack (prior to, or during the attack) and very few fully-developed and deployable tools exist with predictive capability.

Forecasting future events is not a new concept and predictive analytics drives many areas of industry. We are interested in novel approaches to cyber security that can predict the most likely offensive cyber events and/or predict optimal defensive cyber actions, to enable proactive defence in a hostile and contested cyber environment. This competition is anticipated to:

• adapt and implement predictive approaches from other industries to the cyber security domain
• create and implement novel predictive analytics specific to the cyber security domain
• exploit empirical observation-based models of attackers to make predictions (for example of adversary tactics, techniques and procedures; of kill-chains; of attacker competency levels)
• automate the assimilation of (text-based) knowledge collected for many systems (such as known risks or vulnerabilities), and transfer that knowledge to new systems that have the same (or similar) components and operating procedures
• develop approaches to recognise patterns of life that are not time-based, but sequence based
• build on alerts from reactive methods to forecast future offensive cyber events, and thereby predict optimal cyber defences

Proposals that are not in scope include: those that focus on theoretical models, or that lack implementation to real data, and those that ingest social media feeds or other public data of a personal nature.

Predicting vulnerabilities in hardware/software, and monitoring the `health’ of a system are only acceptable if used as components in a larger predictive engine.

Proactive intelligence gathering via the use of honeypots is in scope. Proposals that make use of open-source data formats (for example, threat intelligence reporting, sharing and ingesting) are encouraged. Preference may be given to proposals that forecast future events, rather than predict past events that were overlooked.

We seek to promote collaboration between academia and industry to develop novel tools to prediction in the cyber security domain. All proposals should highlight how subsequent phases will build on the initial phase of development and all phases should include a demonstration as a deliverable. The initial phase may make use of data from enterprise systems (such as standard office equipment) but subsequent phases should show capability when using data from military operational technology.

The initial phase may be demonstrated within a representative business enterprise system but subsequent phases should be applicable to the unique systems, circumstances, threats and opportunities that MOD faces.

2. Funding

Up to £1 million is available for Phase 1 of this competition. We anticipate funding around 5 to 10 proof-of concept research projects of up to 6 months duration. Additional funding may be available for future phases.

3. Competition close

The competition closes at midday on the 5 November 2018.