Security Engineering. Ross Anderson

driving everything from advertising to politics.

      A related but less visible change is the move to large server farms. Sensitive data have moved from servers in schools, doctors' offices and law firms to cloud service providers. Many people no longer do their writing on word processing software on their laptop but on Google Docs or Office365 (I'm writing this book on Overleaf). This has consequences. Security breaches can happen at a scale no-one would have imagined twenty years ago. Compromises of tens of millions of passwords, or credit cards, have become almost routine. And in 2013, we discovered that fifteen years' worth of UK hospital medical records had been sold to 1200 organisations worldwide without the consent of the patients (who were still identifable via their postcodes and dates of birth).

A real game-changer of the last decade was the Snowden revelations, also in 2013, when over 50,000 Top Secret documents about the NSA's signals intelligence activities were leaked to the press. The scale and intrusiveness of government surveillance surprised even cynical security engineers. It followed on from Stuxnet, where America attacked Iran's nuclear weapons program using malware, and was followed by NotPetya, where a Russian cyberweapon, deployed against the Ukraine, inflicted hundreds of millions of dollars' worth of collateral damage on firms elsewhere. This brings us to the third big change, which is a much better understanding of nation-state security threats. In addition to understanding the capabilities and priorities of western intelligence agencies, we have a reasonably good idea of what the Chinese, the Russians and even the Syrians get up to.

      And where the money is, the crooks follow too. The last decade has also seen the emergence of a cyber-crime ecosystem, with malware writers providing the tools to subvert millions of machines, many of which are used as criminal infrastructure while others are subverted in various ways into defrauding their users. We have a team at Cambridge that studies this, and so do dozens of other research groups worldwide. The rise of cybercrime is changing policing, and other state activity too: cryptocurrencies are not just making it easier to write ransomware, but undermining financial regulation. And then there are non-financial threats from cyber-bullying up through hate speech to election manipulation and videos of rape and murder.

      So online harms now engage all sorts of people from teachers and the police to banks and the military. It is ever more important to measure the costs of these harms, and the effectiveness of the measures we deploy to mitigate them.

      Some of the changes would have really surprised someone who read my book ten years ago and then spent a decade in solitary confinement. For example, the multilevel security industry is moribund, despite being the beneficiary of billions of dollars of US government funding over forty years; the Pentagon's entire information security philosophy – of mandating architectures to stop information flowing downward from Top Secret to Secret to Confidential to Unclassified – has been abandoned as unworkable. While architecture still matters, the emphasis has shifted to ecosystems. Given that bugs are ubiquitous and exploits inevitable, we had better be good at detecting exploits, fixing bugs and recovering from attacks. The game is no longer trusted systems but coordinated disclosure, DevSecOps and resilience.

      What might the future hold? A likely game-changer is that as we put software into safety-critical systems like cars and medical devices, and connect them to the Internet, safety and security engineering are converging. This is leading to real strains; while security engineers fix bugs quickly, safety engineers like to test systems rigorously against standards that change slowly if at all. A wicked problem is how we will patch durable goods. At present, you might get security patches for your phone for three years and your laptop for five; you're expected to buy a new one after that. But cars last for fifteen years on average and if we're suddenly asked to scrap them after five the environmental costs won't be acceptable. So tell me, if you're writing navigation software today in 2020 for a car that will launch in 2023, how will you ensure that you can keep on shipping security patches in 2033, 2043 and 2053? What tools will you choose today?

Finally, there has been a sea change in the political environment. After decades in which political leaders considered technology policy to be for men in anoraks, and generally took the line of least resistance, the reports of Russian interference in the Brexit referendum and the Trump election got their attention. The prospect of losing your job can concentrate the mind wonderfully. The close attention of lawmakers is changing the game, first with tighter general rules such as Europe's General Data Protection Regulation; and second as products that are already regulated for safety, from cars and railway signals to children's toys acquire software and online connectivity, which has led to rules in Europe about how long software has to be maintained.

      The questions the security engineer has to ask today are just the same as a decade ago: what are we seeking to prevent, and will the proposed mechanisms actually work? However, the canvas on which we work is now much broader. Almost all human life is there.

      Ross Anderson

      Cambridge, October 2020

Preface to the Second Edition

      The first edition of Security Engineering was published in May 2001. Since then the world has changed.

      System security was one of Microsoft's lowest priorities then; it's now one of the highest. The volume of malware continues to increase along with the nuisance that it causes. Although a lot of effort has gone into defence – we have seen Windows NT replaced by XP and then Vista, and occasional service packs replaced by monthly security patches – the effort put into attacks has increased far more. People who write viruses no longer do so for fun, but for profit; the last few years have seen the emergence of a criminal economy that supports diverse specialists. Spammers, virus writers, phishermen, money launderers and spies trade busily with each other.

      Cryptography has also moved on. The Advanced Encryption Standard is being embedded into more and more products, and we have some interesting developments on the public-key side of things too. But just as our algorithm problems get solved, so we face a host of implementation issues. Side channels, poorly designed APIs and protocol failures continue to break systems. Applied cryptography is harder than ever to do well.

      Pervasive computing also opens up new challenges. As computers and communications become embedded invisibly everywhere, so problems that used to only afflict ‘proper computers’ crop up in all sorts of other devices too. What does it mean for a thermometer to be secure, or an air-conditioner?

The great diversity of intelligent devices brings with it a great diversity of interests and actors. Security is not just about keeping the bad guys out, but increasingly concerned with tussles for power and control. DRM pits the content and platform industries against consumers, and against each other; accessory control is used to tie printers to their vendors' cartridges, but leads to antitrust lawsuits and government intervention. Security also interacts with safety in applications from cars through utilities to electronic healthcare. The security engineer needs to understand not just crypto and operating systems, but economics and human factors as well.

      And the ubiquity of digital devices means that ‘computer security’ is no longer just a problem for a few systems specialists. Almost all white-collar crime (and much crime of the serious violent sort) now involves computers or mobile phones, so a detective needs to understand computer forensics just as she needs to know how to drive. More and more lawyers, accountants, managers and other people with no formal engineering training are going to have to understand system security in order to do their jobs well.

      The rapid growth of online services, from Google and Facebook to massively multiplayer games, has also changed the world. Bugs in online applications can be fixed rapidly once they're noticed, but the applications get ever more complex and their side-effects harder to predict. We may have a reasonably good idea what it means for an operating system or even a banking service to be secure, but we can't make any such claims for online lifestyles that evolve all the time. We're entering a novel world of evolving socio-technical systems, and that raises profound questions about how the evolution is driven and who is in control.

      The largest changes, however, may be those driven by the tragic events of September 2001 and by our reaction to them. These have