We all know that learning a new craft is hard. We spend a large part of our lives learning how to operate in everyday physics. A large part of this learning comes from observing others, and when others can’t help we learn through trial and error.
In machine learning the process of learning how to deal with the environment is called Reinforcement Learning (RL). By continuous interaction with its environment, an agent learns a policy that enables it to perform better. Observational learning in RL is referred to as Imitation Learning. Both trial and error and imitation learning are hard: environments are not trivial, often you can’t tell the ramifications of an action until far in the future, environments are full of non-determinism and there are no such thing as a correct policy.
So, unlike in supervised and unsupervised learning, it is hard to tell if your decisions are correct. Episodes usually constitute thousands of decisions, and you will only know if you perform well after exploring other options. But experiment is also a hard decision: do you exploit the skill you already have, or try something new and explore the unknown?
Despite all these complexities, RL has managed to achieve incredible performance in a wide variety of tasks from robotics through recommender systems to trading. More impressively, RL agents have achieved superhuman performance in Go and other games, tasks previously believed to be impossible for computers.
Continue reading Reinforcement Learning and Adversarial thinking
Good security and cybercrime research often creates an impact and we want to ensure that impact is positive. This week I will discuss three papers on ethics in computer security research in the run up to next week’s Security and Human Behaviour workshop (SHB). Ethical issues in research using datasets of illicit origin (Thomas, Pastrana, Hutchings, Clayton, Beresford) from IMC 2017, Measuring eWhoring (Pastrana, Hutchings, Thomas, Tapiador) from IMC 2019, and An Ethics Framework for Research into Heterogeneous Systems (Happa, Nurse, Goldsmith, Creese, Williams).
Ethical issues in research using datasets of illicit origin (blog post) came about because in prior work we had noticed that there were ethical complexities to take care of when using data that had “fallen off the back of a lorry” such as the backend databases of hacked booter services that we had used. We took a broad look at existing published guidance to synthesise those issues which particularly apply to using data of illicit origin and we expected to see discussed by researchers:
Continue reading Three paper Thursday: Ethics in security research
One would be hard pressed to find an aspect of life where networks are not present. Interconnections are at the core of complex systems – such as society, or the world economy – allowing us to study and understand their dynamics. Some of the most transformative technologies are based on networks, be they hypertext documents making up the World Wide Web, interconnected networking devices forming the Internet, or the various neural network architectures used in deep learning. Social networks that are formed based on our interactions play a central role in our every day lives; they determine how ideas and knowledge spread and they affect behaviour. This is also true for cybercriminal networks present on underground forums, and social network analysis provides valuable insights to how these communities operate either on the dark web or the surface web.
For today’s post in the series `Three Paper Thursday’, I’ve selected three papers that highlight the valuable information we can learn from studying underground forums if we model them as networks. Network topology and large scale structure provide insights to information flow and interaction patterns. These properties along with discovering central nodes and the roles they play in a given community are useful not only for understanding the dynamics of these networks but for various purposes, such as devising disruption strategies.
Continue reading Three Paper Thursday – Analysing social networks within underground forums
Jurassic Park is often (mistakenly) left out of the hacker movie canon. It clearly demonstrated the risk of an insider attack on control systems (Velociraptor rampage, amongst other tragedies…) nearly a decade ahead of the Maroochy sewage incident, it’s the first film I know of with a digital troll (“ah, ah, ah, you didn’t say the magic word!”), and Samuel L. Jackson correctly assesses the possible consequence of a hard reset (namely, everyone dying), resulting in his legendary “Hold on to your butts”. The quotable mayhem is seeded early in the film, when biotech spy Lewis Dodgson gives a sack of money to InGen’s Dennis Nedry to steal some dino DNA. Dodgson’s caricatured OPSEC (complete with trilby and dark glasses) is mocked by Nedry shouting, “Dodgson! Dodgson! We’ve got Dodgson here! See, nobody cares…” Three decades later, this quote still comes to mind* whenever conventional wisdom doesn’t seem to square with observed reality, and today we’re going to apply it to the oft-maligned world of Industrial Control System (ICS) security.
There is plenty of literature on ICS security pre-2010, but people really sat up and started paying attention when we learned about Stuxnet. Possibly the most upsetting thing about Stuxnet (for security-complacent control system designers like me) was the apparent ease with which the “air gap” was bridged over and over again. Any remaining faith in the air gap was killed by Éireann Leverett’s demonstration (thesis and S4 presentation) that thousands of industrial systems were directly connected to the Internet — no air gap jumping required. Since then, we’ve observed a steady growth in Internet-connected ICS devices, due both to improved search techniques and increasingly-connectable ICS devices. On any given day you can find about 100,000 unique devices speaking industrial protocols on Censys and Shodan. These protocols are largely unauthenticated and unencrypted, allowing an attacker that can speak the protocol to remotely read state, issue commands, and even modify programmable logic without using an actual exploit.
This sounds (and is) bad, and people have (correctly) highlighted its badness on many occasions. The attacks, however, appear to be missing: we are not aware of a single instance of industrial damage initiated via an Internet-connected ICS device. In this Three Paper Thursday we’ll look at papers showing how easy it is to find and contextualise Internet-connected ICS devices, some evidence for lack of malicious interest, and some leading indicators that this happy conclusion (for which we don’t really deserve any credit) may be changing.
*Perhaps because guys of a certain age still laugh and say “Dodson! We’ve got Dodson here!” when they learn my surname. I try to explain that it’s spelt differently, but…
Continue reading Three Paper Thursday: Vulnerabilities! We’ve got vulnerabilities here! … See? Nobody cares.
This is a guest contribution from Daniel Woods.
This coming Monday will mark two years since the General Data Protection Regulation (GDPR) came into effect. It prompted an initial wave of cookie banners that drowned users in assertions like “We value your privacy”. Website owners hoped that collecting user consent would ensure compliance and ward off the lofty fines.
Article 6 of the GDPR describes how organisations can establish a legal basis for processing personal data. Putting aside a selection of `necessary’ reasons for doing so, data processing can only be justified by collecting the user’s consent to “the processing of his or her personal data for one or more specific purposes”. Consequently, obtaining user consent could be the difference between suffering a dizzying fine or not.
The law changed the face of the web and this post considers one aspect of the transition. Consent Management Providers (CMPs) emerged offering solutions for websites to embed. Many of these use a technical standard described in the Transparency and Consent Framework. The standard was developed by the Industry Advertising Body, who proudly claim it is is “the only GDPR consent solution built by the industry for the industry”.
All of the following studies either directly measure websites implementing this standard or explore the theoretical implications of standardising consent. The first paper looks at how the design of consent dialogues shape the consent signal sent by users. The second paper identifies disparities between the privacy preferences communicated via cookie banners and the consent signals stored by the website. The third paper uses coalitional game theory to explore which firms extract the value from consent coalitions in which websites share consent signals.
Continue reading Three Paper Thursday – GDPR anniversary edition
Academia, governments and industry frequently talk about the importance of IoT security. Fundamentally, the IoT environment has similar problems to other technology platforms such as Android: a fragmented market with no clear responsibilities or incentives for vendors to provide regular updates, and consumers for whom its not clear how much (of a premium) they are willing to pay for (“better”) security and privacy.
Just two weeks ago, Belkin announced to shut down one of its cloud services, effectively transforming its several product lines of web cameras into useless bricks. Unlike other end-of-support announcements for IoT devices that (only) mean devices will never see an update again, many Belkin cameras simply refuse to work without the “cloud”. This is particularly disconcerting as many see cloud-based IoT as one possible solution to improve device security by easing the user maintenance effort through remote update capabilities.
In this post, I would like to introduce three papers, each talking about different aspects of IoT security: 1) consumer purchasing behaviour, 2) vendor response, and 3) an assessment of the ever-growing literature on “best-practices” from industrial, governmental, and academic sources.
Continue reading Three Paper Thursday: Will we ever get IoT security right?
Software Guard eXtensions (SGX) represents Intel’s latest foray into trusted computing. Initially intended as a means to secure cloud computation, it has since been employed for DRM and secure key storage in production systems. SGX differs from its competitors such as TrustZone in its focus on reducing the volume of trusted code in its “secure world”. These secure worlds are called enclaves in SGX parlance and are protected from untrusted code by a combination of a memory encryption engine and a set of new CPU instructions to enforce separation.
SGX has been available on almost every Intel processor since the sixth generation of Intel Core (Skylake). It has had a bit of a rocky journey since. Upon its release, there were concerns expressed about SGX being used as a medium for malware dissemination. A proof of concept demonstrating the effectiveness of this was also published. Then there were concerns about the mandatory agreements required to publish enclave code in production. In order to use Intel’s attestation service, a key enabler for any remote application built on SGX, developers have to obtain a commercial license from Intel. This lock-in raised a fair bit of consternation and to date hasn’t been fully addressed by Intel. Then came the vulnerabilities. SGX was found to be vulnerable to cache timing attacks, speculative execution vulnerabilities and to Load Value Injection attacks. This has further eroded the credibility of the platform.
That said, there have been a fair few projects that have utilized SGX either to harden existing solutions or to come up with hitherto unseen applications. In this post, we’ll look at three such proposals to see how they utilize SGX despite its concerning attributes. Where applicable, we will also talk about how the discovered vulnerabilities have affected their viability.
Continue reading Three Paper Thursday: What’s Intel SGX Good For?
Just as in other types of victimization, victims of cybercrime can experience serious consequences, emotional or not. First of all, a repeat victim of a cyber-attack might face serious financial or emotional hardship. These victims are also more likely to require medical attention as a consequence of online fraud victimization. This means repeat victims have a unique set of support needs, including the need for counselling, and seeking support from the criminal justice system. There are also cases, such as in cyberbullying or sextortion, where victims will not speak to their family and friends. These victims feel too ashamed to share details with others and they will probably not receive any support. In such cases trauma can even lead to self-harm. Therefore, we see that online victimization can actually lead to physical harm.
As a member of the National Risk Assessment (NRA) Behavioural Science Expert Group in the UK, working on the social and psychological impact of cyber-attacks on members of the public, I have identified for years now that the actual social or psychological impact of different types of cyber-attacks to victims or society as a whole is still not explored. Governments have been slow in identifying and analysing potential events online that may negatively impact individuals. In the UK, as well as in other countries, cybercrime has been added as part of a national risk assessment exercise only a few years ago. Therefore, our knowledge about the potential impact of cyber-attacks and their cascading effects are still being under research.
This is often a very difficult area for lawyers and the courts to understand. Understanding victims’ needs and the responsibilities of the police, the judiciary and other authorities in dealing with such crimes is very important. This is why we need to further explore how and to what extent the situation and needs of victims of online crimes differ from those of traditional offline crimes. By sharing experiences and openly discussing about this issue, we will be able to engrain the cybersecurity mindset in our societies thus preventing victimization in some level.
In this post I would like to introduce recent work in this area. The first one explores the social and psychological impact of cyber-attacks to individuals as well as nations, the second one explores the differences between the situation and needs of online and offline crime victims while the third one discusses the relationship between offending and victimization online.
Continue reading Three Paper Thursday: Exploring the Impact of Online Crime Victimization
People have tried to develop many different attack vectors on cryptocurrencies, from codebase flaws, cryptographic algorithms, mining processes, consensus protocols and block propagation mechanisms to the underlying network layer. Most attacks could be patched quickly by modifying the source code, but preventing attacks that exploit the network layer remains a non-trivial problem as the network layer heavily relies on the existing Internet infrastructure, which is impractical to change. So network-layer attacks could be dangerous, powerful and hard to mitigate.
In this post, I would like to introduce some recent attacks against the Bitcoin P2P network.
Continue reading Three Paper Thursday: Attacking the Bitcoin Peer-to-Peer Network
Recent advancements in Machine Learning (ML) have taught us two main lessons: a large proportion of things that humans do can actually be automated, and that a substantial part of this automation can be done with minimal human supervision. One no longer needs to select features for models to use; in many cases people are moving away from selecting the models themselves and perform a Network Architecture Search. This means non-stop search across billions of dimensions, ever improving different properties of deep neural networks (DNNs).
However, progress in automation has brought a spectre to the feast. Automated systems seem to be very vulnerable to adversarial attacks. Not only is this vulnerability hard to get rid of; worse, we often can’t even define what it means to be vulnerable in the first place.
Furthermore, finding adversarial attacks on ML systems is really easy even if you do not have any access to the models. There are only so many things that make cat a cat, and all the different models that deal with cats will be looking at the same set of features. This has an important implication: learning how to trick one model dealing with cats often transfers over to other models. Transferability is a terrible property for security because it makes adversarial ML attacks cheap and scalable. If there is a camera in the bank running a similar ML model to the camera you can get in Costco for $5, then the cost of developing an attack is $5.
As of now, we do not really have good answers to any of these questions. In the meantime, ML controlled systems are entering the human realm.
In this Three Paper Thursday I want to talk about works from the field of adversarial ML that make it much more understandable.
Continue reading Three Paper Thursday: Adversarial Machine Learning, Humans and everything in between