Slashdot Mirror

Regular Slashdot contributor Bennett Haselton writes in to say "The December 1st release of Psiphon has sparked renewed interest in the various software programs that can help circumvent Internet censorship in China, Iran, and other censored countries. (Some of this interest undoubtedly being motivated by the fact that many of these programs also work for getting around blocking software at work or school.) Have you ever wanted to understand the science behind these programs, the way that mathematicians and codebreakers understand the magic behind PGP? If you loved the mental workout of reading "Applied Cryptography", have you ever wanted a tutorial to do the same for Psiphon and Tor and other anti-censorship programs?" The rest of his editorial follows.

Well, here's a primer, but you might be disappointed. Like making the Statue of Liberty disappear, it doesn't sound very cool once you know how it's done; the truth is that most anti-censorship programs, including mine, only work because the censors are not trying very hard.

(Note that I am going to be talking about ways that certain anti-censorship programs can be defeated. I don't believe that this is giving much help to censors, because these are obvious weaknesses that would occur to anyone who knows how the programs work. For reasons I'll get into at the end, I don't think these weaknesses actually make much difference.)

Basically, all anti-censorship programs fall into two categories: those that require you to have a helper outside of the censored country, and those that don't.

Take Psiphon. To use Psiphon, someone in a non-censored country has to install it on their home computer, which turns their computer into a Web server with an interface similar to Anonymouse.org, where you type in the URL of the page you want to view and it fetches it for you. The difference, of course, is that Anonymouse.org is widely known and blocked by any self-respecting Internet filtering system, while your newly created Psiphon URL pointing to your home computer is not blocked anywhere, yet. So if you set up a Psiphon URL on your computer in the U.S. and e-mail it to your friend in China, your friend can use it to surf wherever they want. (Note that this also has the desirable property that the person in China doesn't have to install any software, so they can use the URL even from a cybercafe computer with restricted user permissions.) The hurdle, of course, is that the person in China has to have a contact outside the country to help them. This is not a huge barrier for many Chinese, but it still means the program doesn't have the instant gratification property of something that you turn on and it just works.

Peacefire, by the way, had released the Circumventor program in 2003 which did essentially the same thing. (And the Circumventor was itself really just a wizard for installing a Web server with James Marshall's CGIProxy script, which deserves most of the credit, although the Circumventor did help bring it "to the masses", since most users don't have the ability to set up an SSL-enabled Web server themselves.) Psiphon made some improvements, namely:

• Ability to create password-protected accounts to restrict the URL to certain users.
• Smaller download (although it may not matter much since only broadband users would be installing it anyway).
• Ability to run on Linux. (Circumventor only works on Windows, although you can install CGIProxy on a Linux webserver if you know how.)

Circumventor has some of its own advantages, although they're the kind that could easily be incorporated into Psiphon soon:

• A wizard to help users forward incoming connections on their router and enter exceptions in software firewalls to make the software work. (If they want to. No tweaking people's firewall settings without asking them!)
• Slightly harder to block, due to some strategies such as using a different SSL certificate for each install (Psiphon uses the same one each time).

And both programs fall victim to the same attacks, although as far as I know, none of these have been implemented in practice:

• Blocking sites whose SSL certificates do not match the site hostname (easier for a censoring proxy server like the ones used in the Middle East, than for an IP firewall like the Great Firewall of China).
• Blocking outgoing Web connections to residential IP address ranges like Comcast.

But basically, they're the same program -- so the difference in press coverage has been illustrative of how much context matters to reporters. Psiphon is the "politically correct" version -- they've played down the fact that it can be used to get around blocking software in schools and played up the fact that it can be used to beat the censors in China and Iran, and the press coverage has focused exclusively on that human rights aspect. The Circumventor was also written to help foreign victims of censorship, and articles have been written about its uses for that purpose, but I've also been unapologetically promoting its use to get around blocking software at home and in school, as part of an advocacy for greater civil rights for people under 18. (Also because the more installations there are in the U.S., the more it helps users abroad.) As a result, some of the TV news pieces about it have used such ominous music and lighting that they practically looked like recycled footage from "To Catch a Predator". Of course, Psiphon can be used for exactly the same thing. (I also emailed some of the reporters who recently wrote about Psiphon, to tell them about Circumventor; so far, I haven't heard back from any of them, but I doubt they're being politically correct this time, I think they're just not thrilled that C-Net scooped them by three years and seven months.)

So, Psiphon and Circumventor fall in the first category -- programs that only work if you've got a contact outside the censored country to help you. In the second category is Tor, which was originally written to provide mathematically secure anonymity, but had the nice property that it could be used to get around the Great Firewall of China as well. With your browser in China using Tor as a proxy, packets are routed to other Tor nodes outside the country, which connect you with any blocked Web site that you want to see. Best of all, you just install it on a machine in China, and presto, it works, no nagging your expat cousin in the U.S. to install something on their computer to help you. Dynamic Internet Technologies, run by Chinese dissident Bill Xia in North Carolina, runs another service that works "out of the box" -- you send an instant-message to one of the DIT screen names, and it replies with a list of currently running Web proxies. (Bill has asked me not to publicize the actual screen names that perform this service, because it's intended only for Chinese users. I think that's a case of "security through obscurity", but I respect his wishes.)

Unfortunately, all such "instant gratification" solutions have the same basic weakness, which by a simple argument can be extended even to hypothetical future programs in the same category. In the case of a program like Tor, the censor only has to install the software, look at what IP addresses the software connects to when it bootstraps itself, and add those IP addresses to the blacklist. Even if the software chooses at random from multiple IP addresses to bootstrap to, the censor can still obtain all of them by repeatedly re-installing the software (possibly wiping the machine each time so the software can't tell that it's been installed before). No matter how you slice it, if Alice the legitimate user and Bob the censor download the program on the same day, Bob can make the program not work for Alice if he updates the blacklist quickly enough. He doesn't even have to reverse-engineer the software, he just has to use a network sniffer to see where it connects to. (For DIT's proxy-by-instant-message system, the censor can instant-message the screen name repeatedly, from different accounts, until they've collected and blocked all the available proxies; this would be analogous to re-installing Tor repeatedly and seeing what IPs it connects to.)

Peacefire has produced other approach which is a simple, obvious idea, and it was quite by accident that we found out it slips through the cracks of the seemingly "unsolvable" problem with instant-gratification outlined above. Like the other solutions, it works only as long as the censors are fairly lazy, but they are, and it does. About 30,000 people have signed up through a form on our site to be notified each time we create a new Circumventor site and mail it out, every 3 or 4 days. Agents of the blocking companies have joined the list too, of course, but we mail different sites to different subsets of the list. Now, an attack analogous to the attacks listed in the previous paragraph, would be for the censors to join under many different accounts, and then block any site that gets mailed to any of those accounts. But the catch is that when an address joins the list, a new site doesn't get mailed to that address until some random time in the future. So the censor has to check all of the fake Hotmail accounts that they've created, over and over, if they want to block all of the new sites as soon as they're released. Hardly impossible, but the censor can no longer use the instantaneous approach of: (1) enter the system / join the list / install the software; (2) see where it connects to and block those points of access; (3) repeat. (If we instantly e-mailed a randomly selected site to each new signup, then this attack would work.) By going from instant gratification to almost-instant-gratification, you change one of the conditions for the theorem stated in the previous paragraph, so that it no longer holds true. Still, like Tor and the DIT system, it could be blocked with a moderate amount of effort.

The Tor protocol, by the way, has been the subject of a great deal of sophisticated mathematical analysis, really brainy stuff that is beyond the scope of this article. But it's important to understand that that analysis focuses on the security of the Tor protocol for achieving anonymity. For anonymity, the protocol is very strong; for routing around censorship, it's fairly straightforward to defeat. That's not at all a criticism of the Tor developers; Tor was designed to achieve anonymity, and just turned out to work for beating censorship as well -- but only, of course, as long as the censors aren't making much effort to block it.

Which all leads to the obvious question: Why have the censors not bothered?

Nobody knows for sure, but I fear the answer is that the Chinese government and other censors know that the greatest weapon in their arsenal is not IP blocking, or keyword filtering, or even the threat of arrest. It's just apathy. The Chinese censors know what we anti-censorware developers in the free world keep forgetting: that most Chinese are not liberty-minded Jeffersonians chomping at the bit under the oppressive yoke of their government and waiting to be freed by circumvention software. As Michael Chase and James Mulvenon of the RAND Corporation put it in their report on Internet usage by Chinese dissidents, You've Got Dissent!: "[A]lthough some peer-to-peer applications... are designed specifically to combat censorship on the Internet and address privacy concerns, most Chinese Internet users are undoubtedly more interested in using peer-to-peer applications for entertainment purposes such as downloading MP3 music files." The censors know what Netscape knew when they fought tooth and nail against Microsoft including Internet Explorer on the desktop of every Windows machine: defaults matter. It doesn't matter that users can go to Netscape's site and download their browser, and it doesn't matter that users can access a banned site by installing a cool p2p program. Most people just don't.

When I first started working on the Circumventor, I assumed that since the Chinese Internet censorship bureau reportedly employed about 30,000 people, surely if they were already spending that much effort and money, they'd throw plenty of resources at defeating any new anti-censorship program, so the Circumventor would have to be able to withstand any such attack. But I was wrong. According to the RAND corporation paper, the censors have been quite busy, for example, policing political forums for dissident postings that other users might casually run into. But they apparently assume -- correctly, it seems -- that content doesn't pose much of a threat if users have to go out of their way and download a program to access it. And if the user has to have a friend outside the country to help them, then forget it.

This is not to downplay the enormous good that programs like Tor, Circumventor and Psiphon can do in bringing free speech to the people in censored countries who want it. But it's easy to forget that those often do not comprise a large part of the population.

One of the biggest disappointments for me came in May 2005 when I was looking for ways to get around the word filter on MSN China's blogging service. Microsoft, apparently acting on public relations advice from Lex Luthor, had decided to filter the words "freedom", "democracy", and "Taiwan independence" from the titles of blogs on MSN China. (I know, I know, they have to comply with Chinese laws to do business there. But I don't think the Chinese have actually outlawed the word "democracy".) Eventually I did find a loophole, so I searched on MSN for some Chinese blogs published by expatriates to ask them to help test the workaround for me. With a few exceptions, most of the bloggers were rather hostile, saying that they supported their government's efforts to censor the Internet and to stamp out Falun Gong as a dangerous "cult". (These were expats living in the U.S., so presumably they were not worried about the Chinese government sending a tank across the Pacific to run them over if they criticized the ruling party. Even if they thought they had to watch what they said because they might someday return to China, or because they still had family there, surely it would have been easier just to ignore me; the hostility that I encountered sounded genuine.) The moral is, no matter how much your movement believes in its efforts to help oppressed people, you can't just assume you'll be greeted as liberators (ahem).

So now you know most of what there is to know about the state of the art in anti-censorship software. It's just that there is less to understand than the hype originally suggests -- the programs aren't really secure, but they work because the censors aren't really trying. And there aren't any cool mathematical formulas that you can impress your friends with -- for that, you'll still have to go back to Applied Cryptography. It's a lot less impressive to be the Bruce Schneier of circumvention algorithms than it is to be the real Bruce Schneier.

Bennett Haselton has another article on offer for us today, this time looking at the implications of a Canadian initiative to protect children online. Bennet writes: "Cybertip.ca, a Canadian clearinghouse for providing information to law enforcement about online child luring and child pornography, has announced that a group of major ISPs will begin blocking access to URLs on Cybertip's list of known child pornography sites. A Cybertip spokesperson says that the list fluctuates between 500 and 800 sites at any given time." Read on for the rest of his analysis. The system is named after a similar filtering system used by service provider BT in the UK. It is also reminiscent of a law passed in Pennsylvania in 2002 requiring ISPs to block URLs on a list of known child pornography sites; the law was struck down in 2004 on First Amendment grounds. Although child pornography is of course not protected by the First Amendment, the law was struck down partly because the ISPs were blocking entire servers and IP address ranges, hundreds of thousands of non-child-pornography sites were also being blocked.

Under the implementation of the Cleanfeed system, representatives from Sasktel, Bell Canada, and Telus claim that only exact URLs will be filtered, not sites hosted at the same IP address. (Although conventional Internet filtering programs sold to parents and schools have also made the same claims, only to turn out to be filtering sites by IP address after all, so we'll have to wait until the filtering is implemented before we know for sure.) The other difference of course is that the Cleanfeed system is not the law, so there's nothing to "strike down" in court. Cybertip did acknowledge that this means customers can get around the filtering for now by switching to a non-participating service provider, although they are encouraging more providers to sign up. Cybertip declined to say whether any providers had simply refused to participate. But of course it's much easier than that to get around the filter, since filter circumvention sites like Anonymouse and StupidCensorship will not be blocked.

So, if it's that easy to circumvent, does it do any good? Even respected Canadian academic and columnist Michael Geist, hardly a friend of censorship in other forms, has spoken out in favor of the plan. I'm going to go out on a limb and say that it doesn't accomplish anything meaningful, and may set a horrible precedent that could make it much easier to block other content in the future.

First of all, it seems that it obviously won't stop anyone who is deliberately looking for child porn. Empirically there's no way to tell -- we don't whether systems like Cleanfeed in the UK have prevented people from accessing child pornography on purpose. Even if the providers are counting the number of blocked accesses to known child porn sites, nobody knows what people have been looking at instead through proxy sites like Anonymouse. All we can do is ask, logically, whether it is likely to work. I think purely logical arguments are frustrating when there is no empirical data to act as a referee, but let's face it, users are not going to self-report on their success at finding child pornography, and there's no way to see what users are accessing through encrypted circumvention sites. Logic is all we have.

So, consider people who are deliberately looking for child pornography. Such people are likely to be resourceful to begin with (since real child porn -- remember, non-sexual pictures of naked children do not count -- is vastly less common than regular porn; Cybertip claims after all that they "only" have about 800 sites on their list, compared to millions of regular porn sites). Virtually all such people would be aware of circumvention sites like Anonymouse, or of peer-to-peer networks, which Cybertip says they have no plans to block. So nothing is blocked from people who want to get around the filter.

The only scenario where the filters could make a difference is the case where someone accidentally accesses a child porn site. Now when I first read the Cybertip press release announcing that the filter would aim to stop "accidental" exposure to child porn, I thought that was just a tactfully sarcastic way of referring to the people who get caught accessing child porn and claim it was just a mistake. But Cybertip.ca claims they've received over 10,000 reports since January 2005 from people who accessed child porn by accident. Even though that only works out to about 15 per day, I have to concede in those cases it almost certainly was a bona fide mistake, for the simple reason that nobody would voluntarily report accessing a child pornography URL that they visited on purpose. But even so, there's the question: What have you accomplished by blocking accidental exposure?

I would argue that the harm done by child pornography is to the minors coerced into the production of it, not to the people who view it. (This, by the way, corresponds with current U.S. jurisprudence; the U.S. Supreme Court ruled in 2002 that a law banning fake child porn was unconstitutional, even when the viewer can't tell the difference.) Obviously you prevent the most damage by stopping child porn at the production stage, but if it's too late for that, you can try to stop people from obtaining it willfully. This lowers the demand and decreases the incentive for people to produce more in the future.

But how would it lower demand if you block people from accessing it accidentally? If those people weren't going to proceed to buy or download more pictures anyway, then they're not fueling the demand. You can block them from accessing the pictures, but the pictures are still out there, and the people who really are fueling the demand can still access them.

So it seems that by blocking someone from accidentally viewing child porn, all you've really accomplished is to avoid offending their sensibilities. Now I don't mean that mockingly, I'm certainly not disagreeing with anyone whose sensibilities are offended by child porn. But there are lots of graphic pictures on the Internet that could offend someone's sensibilities, which are outside of Cleanfeed's mandate. Consider a photo of a 16-year-old having sex, versus a photo of an adult woman fellating a horse; even though the former is illegal to possess and the latter isn't, I think most people would be more grossed out by the second one. (I would even argue that there was more harm to the participants in the making of the second one, and in this case the law's priorities are a bit screwed up. Poor horse!)

So, why block 1% of the content that would offend someone's sensibilities, when 99% of the content that would still offend that person would still be out there? The fact that the 1% is illegal doesn't answer the question; even if it's illegal, you don't have to block it, so what have you accomplished if you do?

Possibly law enforcement is sick of people using the "I accidentally clicked on it" excuse when they get caught accessing child pornography, and wants to remove that as a defense. But couldn't someone just as easily claim that they "accidentally" accessed child pornography through a circumvention site like Anonymouse? They could claim that they thought they were accessing a regular porn site, they were using a circumventor to protect their privacy, and they didn't know that the site carried child porn and didn't find out until they'd already accessed it. So it doesn't seem like the filtering would remove the "accidental" defense.

So, I don't think the filtering accomplishes much at all, but it could set a very bad precedent once the filters are in place. Once Internet users have accepted the precedent that ISPs should block content that is "probably" illegal, what's to stop organizations and lawmakers from demanding that ISPs block access to overseas sites that violate copyright, for example, as the RIAA did in 2002? The technical means will already be in place, and more importantly, people will have gotten used to the idea that legally "questionable" content should be blocked. And with lobbyists claiming that 90% of content on peer-to-peer networks violates copyright laws, wouldn't it follow logically to block peer-to-peer traffic as well?

In a legislative climate where lawmakers have proposed everything from jail time for p2p developers to letting the RIAA hack people's PCs for distributing copyrighted files, we should resist any kind of content-based blocking that would let them get their foot in the door. That includes even well-intentioned efforts like Cleanfeed.