Entries Tagged "passwords"

Page 24 of 29

A Security Market for Lemons

More than a year ago, I wrote about the increasing risks of data loss because more and more data fits in smaller and smaller packages. Today I use a 4-GB USB memory stick for backup while I am traveling. I like the convenience, but if I lose the tiny thing I risk all my data.

Encryption is the obvious solution for this problem—I use PGPdisk—but Secustick sounds even better: It automatically erases itself after a set number of bad password attempts. The company makes a bunch of other impressive claims: The product was commissioned, and eventually approved, by the French intelligence service; it is used by many militaries and banks; its technology is revolutionary.

Unfortunately, the only impressive aspect of Secustick is its hubris, which was revealed when Tweakers.net completely broke its security. There’s no data self-destruct feature. The password protection can easily be bypassed. The data isn’t even encrypted. As a secure storage device, Secustick is pretty useless.

On the surface, this is just another snake-oil security story. But there’s a deeper question: Why are there so many bad security products out there? It’s not just that designing good security is hard—although it is—and it’s not just that anyone can design a security product that he himself cannot break. Why do mediocre security products beat the good ones in the marketplace?

In 1970, American economist George Akerlof wrote a paper called “The Market for ‘Lemons‘” (abstract and article for pay here), which established asymmetrical information theory. He eventually won a Nobel Prize for his work, which looks at markets where the seller knows a lot more about the product than the buyer.

Akerlof illustrated his ideas with a used car market. A used car market includes both good cars and lousy ones (lemons). The seller knows which is which, but the buyer can’t tell the difference—at least until he’s made his purchase. I’ll spare you the math, but what ends up happening is that the buyer bases his purchase price on the value of a used car of average quality.

This means that the best cars don’t get sold; their prices are too high. Which means that the owners of these best cars don’t put their cars on the market. And then this starts spiraling. The removal of the good cars from the market reduces the average price buyers are willing to pay, and then the very good cars no longer sell, and disappear from the market. And then the good cars, and so on until only the lemons are left.

In a market where the seller has more information about the product than the buyer, bad products can drive the good ones out of the market.

The computer security market has a lot of the same characteristics of Akerlof’s lemons market. Take the market for encrypted USB memory sticks. Several companies make encrypted USB drives—Kingston Technology sent me one in the mail a few days ago—but even I couldn’t tell you if Kingston’s offering is better than Secustick. Or if it’s better than any other encrypted USB drives. They use the same encryption algorithms. They make the same security claims. And if I can’t tell the difference, most consumers won’t be able to either.

Of course, it’s more expensive to make an actually secure USB drive. Good security design takes time, and necessarily means limiting functionality. Good security testing takes even more time, especially if the product is any good. This means the less-secure product will be cheaper, sooner to market and have more features. In this market, the more-secure USB drive is going to lose out.

I see this kind of thing happening over and over in computer security. In the late 1980s and early 1990s, there were more than a hundred competing firewall products. The few that “won” weren’t the most secure firewalls; they were the ones that were easy to set up, easy to use and didn’t annoy users too much. Because buyers couldn’t base their buying decision on the relative security merits, they based them on these other criteria. The intrusion detection system, or IDS, market evolved the same way, and before that the antivirus market. The few products that succeeded weren’t the most secure, because buyers couldn’t tell the difference.

How do you solve this? You need what economists call a “signal,” a way for buyers to tell the difference. Warranties are a common signal. Alternatively, an independent auto mechanic can tell good cars from lemons, and a buyer can hire his expertise. The Secustick story demonstrates this. If there is a consumer advocate group that has the expertise to evaluate different products, then the lemons can be exposed.

Secustick, for one, seems to have been withdrawn from sale.

But security testing is both expensive and slow, and it just isn’t possible for an independent lab to test everything. Unfortunately, the exposure of Secustick is an exception. It was a simple product, and easily exposed once someone bothered to look. A complex software product—a firewall, an IDS—is very hard to test well. And, of course, by the time you have tested it, the vendor has a new version on the market.

In reality, we have to rely on a variety of mediocre signals to differentiate the good security products from the bad. Standardization is one signal. The widely used AES encryption standard has reduced, although not eliminated, the number of lousy encryption algorithms on the market. Reputation is a more common signal; we choose security products based on the reputation of the company selling them, the reputation of some security wizard associated with them, magazine reviews, recommendations from colleagues or general buzz in the media.

All these signals have their problems. Even product reviews, which should be as comprehensive as the Tweakers’ Secustick review, rarely are. Many firewall comparison reviews focus on things the reviewers can easily measure, like packets per second, rather than how secure the products are. In IDS comparisons, you can find the same bogus “number of signatures” comparison. Buyers lap that stuff up; in the absence of deep understanding, they happily accept shallow data.

With so many mediocre security products on the market, and the difficulty of coming up with a strong quality signal, vendors don’t have strong incentives to invest in developing good products. And the vendors that do tend to die a quiet and lonely death.

This essay originally appeared in Wired.

EDITED TO ADD (4/22): Slashdot thread.

Posted on April 19, 2007 at 7:59 AM • View Comments

Bank Botches Two-Factor Authentication

From their press release:

The computer was protected by two layers of security, a unique user-identifier and a multiple-character, alpha-numeric password.

Um, hello? Having a username and a password—even if they’re both secret—does not count as two factors, two layers, or two of anything. You need to have two different authentication systems: a password and a biometric, a password and a token.

I wouldn’t trust the New Horizons Community Credit Union with my money.

Posted on April 13, 2007 at 7:33 AM • View Comments

Choosing Secure Passwords

Ever since I wrote about the 34,000 MySpace passwords I analyzed, people have been asking how to choose secure passwords.

My piece aside, there’s been a lot written on this topic over the years—both serious and humorous—but most of it seems to be based on anecdotal suggestions rather than actual analytic evidence. What follows is some serious advice.

The attack I’m evaluating against is an offline password-guessing attack. This attack assumes that the attacker either has a copy of your encrypted document, or a server’s encrypted password file, and can try passwords as fast as he can. There are instances where this attack doesn’t make sense. ATM cards, for example, are secure even though they only have a four-digit PIN, because you can’t do offline password guessing. And the police are more likely to get a warrant for your Hotmail account than to bother trying to crack your e-mail password. Your encryption program’s key-escrow system is almost certainly more vulnerable than your password, as is any “secret question” you’ve set up in case you forget your password.

Offline password guessers have gotten both fast and smart. AccessData sells Password Recovery Toolkit, or PRTK. Depending on the software it’s attacking, PRTK can test up to hundreds of thousands of passwords per second, and it tests more common passwords sooner than obscure ones.

So the security of your password depends on two things: any details of the software that slow down password guessing, and in what order programs like PRTK guess different passwords.

Some software includes routines deliberately designed to slow down password guessing. Good encryption software doesn’t use your password as the encryption key; there’s a process that converts your password into the encryption key. And the software can make this process as slow as it wants.

The results are all over the map. Microsoft Office, for example, has a simple password-to-key conversion, so PRTK can test 350,000 Microsoft Word passwords per second on a 3-GHz Pentium 4, which is a reasonably current benchmark computer. WinZip used to be even worse—well over a million guesses per second for version 7.0—but with version 9.0, the cryptosystem’s ramp-up function has been substantially increased: PRTK can only test 900 passwords per second. PGP also makes things deliberately hard for programs like PRTK, also only allowing about 900 guesses per second.

When attacking programs with deliberately slow ramp-ups, it’s important to make every guess count. A simple six-character lowercase exhaustive character attack, “aaaaaa” through “zzzzzz,” has more than 308 million combinations. And it’s generally unproductive, because the program spends most of its time testing improbable passwords like “pqzrwj.”

According to Eric Thompson of AccessData, a typical password consists of a root plus an appendage. A root isn’t necessarily a dictionary word, but it’s something pronounceable. An appendage is either a suffix (90 percent of the time) or a prefix (10 percent of the time).

So the first attack PRTK performs is to test a dictionary of about 1,000 common passwords, things like “letmein,” “password,” “123456” and so on. Then it tests them each with about 100 common suffix appendages: “1,” “4u,” “69,” “abc,” “!” and so on. Believe it or not, it recovers about 24 percent of all passwords with these 100,000 combinations.

Then, PRTK goes through a series of increasingly complex root dictionaries and appendage dictionaries. The root dictionaries include:

  • Common word dictionary: 5,000 entries
  • Names dictionary: 10,000 entries
  • Comprehensive dictionary: 100,000 entries
  • Phonetic pattern dictionary: 1/10,000 of an exhaustive character search

The phonetic pattern dictionary is interesting. It’s not really a dictionary; it’s a Markov-chain routine that generates pronounceable English-language strings of a given length. For example, PRTK can generate and test a dictionary of very pronounceable six-character strings, or just-barely pronounceable seven-character strings. They’re working on generation routines for other languages.

PRTK also runs a four-character-string exhaustive search. It runs the dictionaries with lowercase (the most common), initial uppercase (the second most common), all uppercase and final uppercase. It runs the dictionaries with common substitutions: “$” for “s,” “@” for “a,” “1” for “l” and so on. Anything that’s “leet speak” is included here, like “3” for “e.”

The appendage dictionaries include things like:

  • All two-digit combinations
  • All dates from 1900 to 2006
  • All three-digit combinations
  • All single symbols
  • All single digit, plus single symbol
  • All two-symbol combinations

AccessData’s secret sauce is the order in which it runs the various root and appendage dictionary combinations. The company’s research indicates that the password sweet spot is a seven- to nine-character root plus a common appendage, and that it’s much more likely for someone to choose a hard-to-guess root than an uncommon appendage.

Normally, PRTK runs on a network of computers. Password guessing is a trivially distributable task, and it can easily run in the background. A large organization like the Secret Service can easily have hundreds of computers chugging away at someone’s password. A company called Tableau is building a specialized FPGA hardware add-on to speed up PRTK for slow programs like PGP and WinZip: roughly a 150- to 300-times performance increase.

How good is all of this? Eric Thompson estimates that with a couple of weeks’ to a month’s worth of time, his software breaks 55 percent to 65 percent of all passwords. (This depends, of course, very heavily on the application.) Those results are good, but not great.

But that assumes no biographical data. Whenever it can, AccessData collects whatever personal information it can on the subject before beginning. If it can see other passwords, it can make guesses about what types of passwords the subject uses. How big a root is used? What kind of root? Does he put appendages at the end or the beginning? Does he use substitutions? ZIP codes are common appendages, so those go into the file. So do addresses, names from the address book, other passwords and any other personal information. This data ups PRTK’s success rate a bit, but more importantly it reduces the time from weeks to days or even hours.

So if you want your password to be hard to guess, you should choose something not on any of the root or appendage lists. You should mix upper and lowercase in the middle of your root. You should add numbers and symbols in the middle of your root, not as common substitutions. Or drop your appendage in the middle of your root. Or use two roots with an appendage in the middle.

Even something lower down on PRTK’s dictionary list—the seven-character phonetic pattern dictionary—together with an uncommon appendage, is not going to be guessed. Neither is a password made up of the first letters of a sentence, especially if you throw numbers and symbols in the mix. And yes, these passwords are going to be hard to remember, which is why you should use a program like the free and open-source Password Safe to store them all in. (PRTK can test only 900 Password Safe 3.0 passwords per second.)

Even so, none of this might actually matter. AccessData sells another program, Forensic Toolkit, that, among other things, scans a hard drive for every printable character string. It looks in documents, in the Registry, in e-mail, in swap files, in deleted space on the hard drive … everywhere. And it creates a dictionary from that, and feeds it into PRTK.

And PRTK breaks more than 50 percent of passwords from this dictionary alone.

What’s happening is that the Windows operating system’s memory management leaves data all over the place in the normal course of operations. You’ll type your password into a program, and it gets stored in memory somewhere. Windows swaps the page out to disk, and it becomes the tail end of some file. It gets moved to some far out portion of your hard drive, and there it’ll sit forever. Linux and Mac OS aren’t any better in this regard.

I should point out that none of this has anything to do with the encryption algorithm or the key length. A weak 40-bit algorithm doesn’t make this attack easier, and a strong 256-bit algorithm doesn’t make it harder. These attacks simulate the process of the user entering the password into the computer, so the size of the resultant key is never an issue.

For years, I have said that the easiest way to break a cryptographic product is almost never by breaking the algorithm, that almost invariably there is a programming error that allows you to bypass the mathematics and break the product. A similar thing is going on here. The easiest way to guess a password isn’t to guess it at all, but to exploit the inherent insecurity in the underlying operating system.

This essay originally appeared on Wired.com.

Posted on January 11, 2007 at 8:04 AM • View Comments

Real-World Passwords

How good are the passwords people are choosing to protect their computers and online accounts?

It’s a hard question to answer because data is scarce. But recently, a colleague sent me some spoils from a MySpace phishing attack: 34,000 actual user names and passwords.

The attack was pretty basic. The attackers created a fake MySpace login page, and collected login information when users thought they were accessing their own account on the site. The data was forwarded to various compromised web servers, where the attackers would harvest it later.

MySpace estimates that more than 100,000 people fell for the attack before it was shut down. The data I have is from two different collection points, and was cleaned of the small percentage of people who realized they were responding to a phishing attack. I analyzed the data, and this is what I learned.

Password Length: While 65 percent of passwords contain eight characters or less, 17 percent are made up of six characters or less. The average password is eight characters long.

Specifically, the length distribution looks like this:

1-4 0.82 percent
5 1.1 percent
6 15 percent
7 23 percent
8 25 percent
9 17 percent
10 13 percent
11 2.7 percent
12 0.93 percent
13-32 0.93 percent

Yes, there’s a 32-character password: “1ancheste23nite41ancheste23nite4.” Other long passwords are “fool2thinkfool2thinkol2think” and “dokitty17darling7g7darling7.”

Character Mix: While 81 percent of passwords are alphanumeric, 28 percent are just lowercase letters plus a single final digit—and two-thirds of those have the single digit 1. Only 3.8 percent of passwords are a single dictionary word, and another 12 percent are a single dictionary word plus a final digit—once again, two-thirds of the time that digit is 1.

numbers only 1.3 percent
letters only 9.6 percent
alphanumeric 81 percent
non-alphanumeric 8.3 percent

Only 0.34 percent of users have the user name portion of their e-mail address as their password.

Common Passwords: The top 20 passwords are (in order): password1, abc123, myspace1, password, blink182, qwerty1, fuckyou, 123abc, baseball1, football1, 123456, soccer, monkey1, liverpool1, princess1, jordan23, slipknot1, superman1, iloveyou1 and monkey. (Different analysis here.)

The most common password, “password1,” was used in 0.22 percent of all accounts. The frequency drops off pretty fast after that: “abc123” and “myspace1” were only used in 0.11 percent of all accounts, “soccer” in 0.04 percent and “monkey” in 0.02 percent.

For those who don’t know, Blink 182 is a band. Presumably lots of people use the band’s name because it has numbers in its name, and therefore it seems like a good password. The band Slipknot doesn’t have any numbers in its name, which explains the 1. The password “jordan23” refers to basketball player Michael Jordan and his number. And, of course, “myspace” and “myspace1” are easy-to-remember passwords for a MySpace account. I don’t know what the deal is with monkeys.

We used to quip that “password” is the most common password. Now it’s “password1.” Who said users haven’t learned anything about security?

But seriously, passwords are getting better. I’m impressed that less than 4 percent were dictionary words and that the great majority were at least alphanumeric. Writing in 1989, Daniel Klein was able to crack (.gz) 24 percent of his sample passwords with a small dictionary of just 63,000 words, and found that the average password was 6.4 characters long.

And in 1992 Gene Spafford cracked (.pdf) 20 percent of passwords with his dictionary, and found an average password length of 6.8 characters. (Both studied Unix passwords, with a maximum length at the time of 8 characters.) And they both reported a much greater percentage of all lowercase, and only upper- and lowercase, passwords than emerged in the MySpace data. The concept of choosing good passwords is getting through, at least a little.

On the other hand, the MySpace demographic is pretty young. Another password study (.pdf) in November looked at 200 corporate employee passwords: 20 percent letters only, 78 percent alphanumeric, 2.1 percent with non-alphanumeric characters, and a 7.8-character average length. Better than 15 years ago, but not as good as MySpace users. Kids really are the future.

None of this changes the reality that passwords have outlived their usefulness as a serious security device. Over the years, password crackers have been getting faster and faster. Current commercial products can test tens—even hundreds—of millions of passwords per second. At the same time, there’s a maximum complexity to the passwords average people are willing to memorize (.pdf). Those lines crossed years ago, and typical real-world passwords are now software-guessable. AccessData’s Password Recovery Toolkit—at 200,000 guesses per second—would have been able to crack 23 percent of the MySpace passwords in 30 minutes, 55 percent in 8 hours.

Of course, this analysis assumes that the attacker can get his hands on the encrypted password file and work on it offline, at his leisure; i.e., that the same password was used to encrypt an e-mail, file or hard drive. Passwords can still work if you can prevent offline password-guessing attacks, and watch for online guessing. They’re also fine in low-value security situations, or if you choose really complicated passwords and use something like Password Safe to store them. But otherwise, security by password alone is pretty risky.

This essay originally appeared on Wired.com.

Posted on December 14, 2006 at 7:39 AM • View Comments

Fighting Fraudulent Transactions

Last March I wrote that two-factor authentication isn’t going to reduce financial fraud or identity theft, that all it will do is force the criminals to change their tactics:

Unfortunately, the nature of attacks has changed over those two decades. Back then, the threats were all passive: eavesdropping and offline password guessing. Today, the threats are more active: phishing and Trojan horses.

Here are two new active attacks we’re starting to see:

  • Man-in-the-Middle attack. An attacker puts up a fake bank website and entices user to that website. User types in his password, and the attacker in turn uses it to access the bank’s real website. Done right, the user will never realize that he isn’t at the bank’s website. Then the attacker either disconnects the user and makes any fraudulent transactions he wants, or passes along the user’s banking transactions while making his own transactions at the same time.
  • Trojan attack. Attacker gets Trojan installed on user’s computer. When user logs into his bank’s website, the attacker piggybacks on that session via the Trojan to make any fraudulent transaction he wants.

See how two-factor authentication doesn’t solve anything? In the first case, the attacker can pass the ever-changing part of the password to the bank along with the never-changing part. And in the second case, the attacker is relying on the user to log in.

The solution is not to better authenticate the person, but to authenticate the transaction. (Think credit cards. No one checks your signature. They really don’t care if you’re you. They maintain security by authenticating the transactions.)

Of course, no one listens to me. U.S. regulators required banks to implement two-factor authentication by the end of this year. But customers are rebelling, and banks are scrambling to figure out something—anything—else. And, amazingly enough and purely by accident it seems, they’ve stumbled on security solutions that actually work:

Instead, to comply with new banking regulations and stem phishing losses, banks and the vendors who serve them are hurriedly putting together multipronged strategies that they say amount to “strong” authentication. The emerging approach generally consists of somehow recognizing a customer’s computer, asking additional challenge questions for risky behavior and putting in place back-end fraud detection.

[…]

Despite the FFIEC guidance about authentication, the emerging technologies that actually seem to hold the most promise for protecting the funds in consumer banking accounts aren’t authentication systems at all. They’re back-end systems that monitor for suspicious behavior.

Some of these tools are rule-based: If a customer from Nebraska signs on from, say, Romania, the bank can determine that the log-on always be considered suspect. Others are based on a risk score: That log-on from Romania would add points to a risk score, and when the score reaches a certain threshold, the bank takes action.

Flagged transactions can get bumped to second-factor authentication—usually, a call on the telephone, something the user has. This has long been done manually in the credit card world. Just think about the last phone call you got from your credit card company’s fraud department when you (or someone else) tried to make a large purchase with your credit card in Europe. Some banks, including Washington Mutual, are in the process of automating out-of-band phone calls for risky online transactions.

Exactly. That’s how you do it.

EDITED TO ADD (12/6): Another example.

Posted on November 27, 2006 at 6:07 AM • View Comments

1 … 22 23 24 25 26 … 29

Sidebar photo of Bruce Schneier by Joe MacInnis.