← All posts

AI Expert Says: Cybercrime Solution is Government, Not AI

Cybercrime is growing at an alarming rate, and artificial intelligence is supposed to be the solution. This isn’t good enough. Broader, the private security industry isn’t even enough to solve this problem.

AI produced by private companies will help, but it has major blindspots such as targeted, high-value attacks. Ignoring structural, government-based solutions will lead to more attacks, more cost, and even greater risk to your health.

Our growing cybercrime problem

The FBI just released its annual report on cybercrime,1 with the cost rising to $3.5 billion in 2019. This is up 30% from last year2 and up 720% since the FBI started reporting the cost in 20113.

These growth figures are useful, but this report dramatically understates the cost of cybercrime, only including the small fraction of crimes that are reported to the FBI and leaving out massive indirect costs like money spent on preventative security. The total cost has been estimated between $600 billion4 and several trillion dollars5 annually.

The scale of the attacks is so enormous that humans can’t keep up. As many as 32-45% of tech jobs available in the US are in the field of cybersecurity.6

Artificial intelligence to the rescue

Fortunately, artificial intelligence is going to solve the problem! Or so it might appear when reading the news on this topic.7

This trend of AI becoming a critical component of cybersecurity efforts is real. As someone who worked at Microsoft for over a decade using AI to fight large-scale abuse on the Internet, I certainly understand how critical this technology is.

Consider not using AI to solve cybercrime. Even if there weren’t privacy challenges, it would not be feasible to pay humans to review every piece of content a person could see online and determine whether it’s malicious or not.

There are hundreds of trillions of websites people could stumble onto,8 many generated by machines. And this has historically doubled each year.9 Our inability to manually review them all is not just an issue of cost, though. The collective capacity of every human on Earth is insufficient to keep up.

Fortunately, the most well-developed area of AI is detecting patterns in data. Detecting malicious activity can be automated. Since the criminals and abusers are automating their attacks, security must be automated as well to keep up.

…with limitations

That said, AI is not perfect. It can be very effective at mitigating widespread attacks, such as building a spam filter. In this area, there are massive datasets available. Many people receive similar spam messages as part of the same attack, so patterns in the data are plentiful.

Of course, if you have an email address you also know these systems make mistakes, missing some spam and filtering out some legitimate emails. Some mistakes are manageable in the case of email filtering, but mistakes wouldn’t be acceptable for defending a nuclear launch system.

AI struggles with targeted, high-value attacks. For example, an abuser could spend substantial time personalizing a targeted attack on well-chosen target at a hospital, then use that success to break into the hospital’s network and hold their data hostage in exchange for a ransom.

This is, unfortunately, all too common and can lead to hospitals shutting down when their critical IT systems become useless.10 Cybercrime can even be a risk to your health.

How security firms have responded

Security companies have responded both by trying as hard as they can to secure their systems and prevent attacks—which is generally believed to be impossible. But they’ve also come up with an innovation: learning what good activity looks like and using AI to flag anything anomalous for human review.

Unfortunately, the world complex and lots of weird things happen. System administrators get so many alerts that they can’t possibly respond to all of them, and certainly not promptly.

Administrators can decide to let suspicious activity slide while it awaits review, or they can block suspicious activity and frustrate employees who now constantly need an administrator to put in a password so they can get their work done.

To be clear, artificial intelligence is a critical technology that everyone should be adopting to improve their security. It is unparalleled at stopping massive attacks, and even offers assistance with anomaly detection that leaves us in a much better spot than we’d be without it.

My point is not that AI is a waste of time, but rather that it is not a panacea. Much more than AI will be needed to solve these challenging problems.

The fundamental problem

Imagine being back in elementary school and getting attacked by a bully. As he throws a punch, you hold your arms up to block it. You don’t punch back, tell a teacher, or even run to friends for help.

Even if you block one punch, that’s not going to be the end of it. And after getting beat-up once, without any consequences the bully will just come back and do it again and again.

This is the state of cybersecurity today.

Far less than 1% of cybercrime leads to a successful prosecution.11

There is so much crime online because the criminals perpetrating it are rarely punished. They can safely hide in Russia, China, and many smaller countries where they will not be held accountable for their actions. They can safely extract money from victims on the other side of the world.

When one of the platform companies like Facebook or Google detects some abuse, they prevent their customers from getting tricked by it. The attacker sees the current campaign is now getting detected and launches a new one.

Companies can hold their arms up and protect themselves, but they can’t stop the attacker from coming back. They can’t put the attacker in jail directly, of course. And since local authorities can’t prosecute foreign criminals, there’s no authority able and willing to help in most cases.

Artificial intelligence can enable these companies to hold up dozens of arms at just the right angles to make it really hard to land a punch. But with enough attempts, they’ll still get through. And they can come back in perpetuity.

How do we solve this?

Only governments have the power to solve these problems.

Artificial intelligence and other security features built by companies are essential for protection, but the level of investment necessary would be much less if criminals were held responsible and there was a meaningful deterrent to this crime.

Big tech companies stop an almost unimaginable number of attacks each day, and these mostly go unreported since pursuing them through the legal system is hopeless. They need to keep doing this amazing work to protect their users, even if it doesn’t solve the underlying problem.

The FBI and many other governmental entities around the world do a great job tracking down criminals when they happen to live in countries like the US or Western Europe where cybercriminals are held responsible. This is also great and heroic work which must continue.

Although there are certainly places in government where the technical capabilities could be improved, this is not the issue. The problems that need to be solved are ones squarely within the core competency of government: jurisdiction and foreign relations.

A more detailed account of how this problem can be solved, which I’m actively working on, will have to wait for a future discussion. Nonetheless, this is already sufficient information to draw some troubling conclusions.

Conclusion

Cybercrime is a massive and growing problem. AI is the solution everyone is turning towards. This is helpful, and even an essential technology for mitigating the threat, but it’s fundamentally incapable of solving the root of the problem. Going further, private companies are incapable of solving it.

The real problems behind this steep growth in cybercrime are structural issues that need to be resolved at the governmental level.

References

[1] https://www.fbi.gov/news/stories/2019-internet-crime-report-released-021120

[2] https://www.fbi.gov/news/stories/ic3-releases-2018-internet-crime-report-042219

[3] https://archives.fbi.gov/archives/news/pressrel/press-releases/ic3-2011-internet-crime-report-released

[4] https://www.mcafee.com/enterprise/en-us/solutions/lp/economics-cybercrime.html

[5] https://www.prnewswire.com/news-releases/cyberattacks-are-the-fastest-growing-crime-and-predicted-to-cost-the-world-6-trillion-annually-by-2021-300765090.html

[6] https://cybersecurityventures.com/jobs/

[7] https://www.forbes.com/sites/louiscolumbus/2019/07/14/why-ai-is-the-future-of-cybersecurity/#5a8752b9117e

[8] https://www.brandwatch.com/blog/amazing-social-media-statistics-and-facts/

[9] https://venturebeat.com/2013/03/01/how-google-searches-30-trillion-web-pages-100-billion-times-a-month/

[10] https://threatpost.com/ransomware-attacks-leave-u-s-hospitals-turning-away-patients/148823/

[11] Eoyang, M., et al., To Catch a Hacker: Toward a comprehensive strategy to identify, pursue, and punish malicious cyber actors. 2018, Third Way.