Disclosing vs Hoarding Vulnerabilities

There's a debate going on about whether the US government -- specifically, the NSA and United States Cyber Command -- should stockpile Internet vulnerabilities or disclose and fix them. It's a complicated problem, and one that starkly illustrates the difficulty of separating attack and defense in cyberspace.
A software vulnerability is a programming mistake that allows an adversary access into that system. Heartbleed is a recent example, but hundreds are discovered every year.
Unpublished vulnerabilities are called "zero-day" vulnerabilities, and they're very valuable because no one is protected. Someone with one of those can attack systems world-wide with impunity.
When someone discovers one, he can either use it for defense or for offense. Defense means alerting the vendor and getting it patched. Lots of vulnerabilities are discovered by the vendors themselves and patched without any fanfare. Others are discovered by researchers and hackers. A patch doesn't make the vulnerability go away, but most users protect themselves by patch their systems regularly.
Offense means using the vulnerability to attack others. This is the quintessential zero-day, because the vendor doesn't even know the vulnerability exists until it starts being used by criminals or hackers. Eventually the affected software's vendor finds out -- the timing depends on how extensively the vulnerability is used -- and issues a patch to close the vulnerability.
If an offensive military cyber unit discovers the vulnerability -- or a cyber-weapons arms manufacturer -- it keeps that vulnerability secret for use to deliver a cyber-weapon. If it is used stealthily, it might remain secret for a long time. If unused, it'll remain secret until someone else discovers it.
Discoverers can sell vulnerabilities. There's a rich market in zero-days for attack purposes -- both military/commercial and black markets. Some vendors offer bounties for vulnerabilities to incent defense, but the amounts are much lower.
The NSA can play either defense or offense. It can either alert the vendor and get a still-secret vulnerability fixed, or it can hold on to it and use it as to eavesdrop on foreign computer systems. Both are important US policy goals, but the NSA has to choose which one to pursue. By fixing the vulnerability, it strengthens the security of the Internet against all attackers: other countries, criminals, hackers. By leaving the vulnerability open, it is better able to attack others on the Internet. But each use runs the risk of the target government learning of, and using for itself, the vulnerability -- or of the vulnerability becoming public and criminals starting to use it.
There is no way to simultaneously defend US networks while leaving foreign networks open to attack. Everyone uses the same software, so fixing us means fixing them, and leaving them vulnerable means leaving us vulnerable. As Harvard Law Professor Jack Goldsmith wrote, "every offensive weapon is a (potential) chink in our defense -- and vice versa."
To make matters even more difficult, there is an arms race going on in cyberspace. The Chinese, the Russians, and many other countries are finding vulnerabilities as well. If we leave a vulnerability unpatched, we run the risk of another country independently discovering it and using it in a cyber-weapon that we will be vulnerable to. But if we patch all the vulnerabilities we find, we won't have any cyber-weapons to use against other countries.
Many people have weighed in on this debate. The president's Review Group on Intelligence and Communications Technologies, convened post-Snowden, concluded (recommendation 30), that vulnerabilities should only be hoarded in rare instances and for short times. Cory Doctorow calls it a public health problem.  Dan Geer recommends that the US government corner the vulnerabilities market and fix them all. Both the FBI and the intelligence agencies claim that this amounts to unilateral disarmament.
It seems like an impossible puzzle, but the answer hinges on how vulnerabilities are distributed in software.
If vulnerabilities are sparse, then it's obvious that every vulnerability we find and fix improves security. We render a vulnerability unusable, even if the Chinese government already knows about it. We make it impossible for criminals to find and use it. We improve the general security of our software, because we can find and fix most of the vulnerabilities.
If vulnerabilities are plentiful -- and this seems to be true -- the ones the US finds and the ones the Chinese find will largely be different. This means that patching the vulnerabilities we find won't make it appreciably harder for criminals to find the next one. We don't really improve general software security by disclosing and patching unknown vulnerabilities, because the percentage we find and fix is small compared to the total number that are out there.
But while vulnerabilities are plentiful, they're not uniformly distributed. There are easier-to-find ones, and harder-to-find ones. Tools that automatically find and fix entire classes of vulnerabilities, and coding practices that eliminate many easy-to-find ones, greatly improve software security. And when person finds a vulnerability, it is likely that another person soon will, or recently has, found the same vulnerability. Heartbleed, for example, remained undiscovered for two years, and then two independent researchers discovered it within two days of each other. This is why it is important for the government to err on the side of disclosing and fixing.
The NSA, and by extension US Cyber Command, tries its best to play both ends of this game. Former NSA Director Michael Hayden talks about NOBUS, "nobody but us." The NSA has a classified process to determine what it should do about vulnerabilities, disclosing and closing most of the ones it finds, but holding back some -- we don't know how many -- vulnerabilities that "nobody but us" could find for attack purposes.
This approach seems to be the appropriate general framework, but the devil is in the details. Many of us in the security field don't know how to make NOBUS decisions, and the recent White House clarification posed more questions than it answered.
Who makes these decisions, and how? How often are they reviewed? Does this review process happen inside Department of Defense, or is it broader? Surely there needs to be a technical review of each vulnerability, but there should also be policy reviews regarding the sorts of vulnerabilities we are hoarding. Do we hold these vulnerabilities until someone else finds them, or only for a short period of time? How many do we stockpile? The US/Israeli cyberweapon Stuxnet used four zero-day vulnerabilities. Burning four on a single military operation implies that we are not hoarding a small number, but more like 100 or more.
There's one more interesting wrinkle. Cyber-weapons are a combination of a payload -- the damage the weapon does -- and a delivery mechanism: the vulnerability used to get the payload into the enemy network. Imagine that China knows about a vulnerability and is using it in a still-unfired cyber-weapon, and that the NSA learns about it through espionage. Should the NSA disclose and patch the vulnerability, or should it use it itself for attack? If it discloses, then China could find a replacement vulnerability that the NSA won't know about it. But if it doesn't, it's deliberately leaving the US vulnerable to cyber-attack. Maybe someday we can get to the point where we can patch vulnerabilities faster than the enemy can use them in an attack, but we're nowhere near that point today.
The implications of US policy can be felt on a variety of levels. The NSA's actions have resulted in a widespread mistrust of the security of US Internet products and services, greatly affecting American business. If we show that we're putting security ahead of surveillance, we can begin to restore that trust. And by making the decision process much more public than it is today, we can demonstrate both our trustworthiness and the value of open government.
An unpatched vulnerability puts everyone at risk, but not to the same degree. The US and other Western countries are highly vulnerable, because of our critical electronic infrastructure, intellectual property, and personal wealth. Countries like China and Russia are less vulnerable -- North Korea much less -- so they have considerably less incentive to see vulnerabilities fixed. Fixing vulnerabilities isn't disarmament; it's making our own countries much safer. We also regain the moral authority to negotiate any broad international reductions in cyber-weapons; and we can decide not to use them even if others do.
Regardless of our policy towards hoarding vulnerabilities, the most important thing we can do is patch vulnerabilities quickly once they are disclosed. And that's what companies are doing, even without any government involvement, because so many vulnerabilities are discovered by criminals.
We also need more research in automatically finding and fixing vulnerabilities, and in building secure and resilient software in the first place. Research over the last decade or so has resulted in software vendors being able to find and close entire classes of vulnerabilities. Although there are many cases of these security analysis tools not being used, all of our security is improved when they are. That alone is a good reason to continue disclosing vulnerability details, and something the NSA can do to vastly improve the security of the Internet worldwide. Here again, though, they would have to make the tools they have to automatically find vulnerabilities available for defense and not attack.
In today's cyberwar arms race, unpatched vulnerabilities and stockpiled cyber-weapons are inherently destabilizing, especially because they are only effective for a limited time. The world's militaries are investing more money in finding vulnerabilities than the commercial world is investing in fixing them. The vulnerabilities they discover affect the security of us all. No matter what cybercriminals do, no matter what other countries do, we in the US need to err on the side of security and fix almost all the vulnerabilities we find. But not all, yet.