Silly as they seem, U.S. intelligence agents consider these questions key to their victory in the war on terrorism, according to unnamed sources who have been quoted in media reports over the past year. Since before Sept. 11, a series of articles have quoted experts suggesting that al-Qaida may be especially Internet-savvy and could be mounting a full-scale “cyberwar” against the United States.

While much of it comes off as alarmist speculation, one hard-to-prove fact has slowly gained a patina of credibility: that terrorists are hiding coded messages in the image files on eBay and other sites that allow public posting. These images would appear normal to most eBay shoppers, but they are actually brimming with guile. A terrorist who knew their true purpose could download the files, decode them with his secret password and perhaps find out where to strike next.

Jack Kelley, a veteran foreign correspondent for USA Today, has been at the forefront of these reports. In February 2001, Kelley reported that hidden “in the X-rated pictures on several pornographic Web sites and the posted comments on sports chat rooms may lie the encrypted blueprints of the next terrorist attack against the United States or its allies.”

His report prompted a flurry of follow-up stories in other publications, including one Wired News story in which a security expert said that his company, WetStone Technologies, had found several hidden messages on eBay and Amazon. After Sept. 11, dozens of newspapers, including the New York Times and the Washington Post, cited WetStone in reports that eBay may be crawling with terrorists. These accounts were almost universally dismissed by Internet-rights types, who said that they wouldn’t believe the stories until they saw proof that “steganography” — the practice of digitally hiding messages in media files — is indeed on the rise.

On July 10, USA Today prompted renewed interest in the steganography debate by adding some meat to the eBay story. “Lately, al-Qaida operatives have been sending hundreds of encrypted messages that have been hidden in files on digital photographs on the auction site eBay.com,” reported Jack Kelley. “The volume of the messages has nearly doubled in the past month, indicating to some U.S. intelligence officials that al-Qaida is planning another attack.” Kelley added that eBay did not return his calls for comment.

The USA Today article has raised plenty of eyebrows — eBay for example, has no record of being contacted by Kelley, and stresses that no federal agency has alerted it to any potential problems. There also appears to be little, if any, publicly available hard evidence of the use of steganography in files on the auction site.

The frightful genius of steganography, though, is that, by design, you don’t know when it’s being used. Independent researchers have devised numerous methods to search for signs of its proliferation on the Web, and some have reported that they’ve found nothing, and there’s consequently no reason to be afraid. But when you think about these studies, the results become about as comforting as homeland security advisor Tom Ridge’s color-coded alert system. After all, if you search for hidden messages on the Web and find nothing, what should you conclude — that there are no messages, or that the terrorists are too sophisticated, and your tools don’t work?

The answer to this question turns out to be a highly personal one, a matter of individual psychology and interest rather than a reasoned decision based on collective safety and the immutable laws of math. Ask security types, or people who make software to aid security types, and they say that steganography is a grave threat to our safety. Defenders of steganography, and its cousin cryptography, take the opposite view. These are people who become easily exercised over the prospect of the government monitoring the Web, and they say that if researchers haven’t found secret messages, the messages are likely not there. But amid this politicking, one important question tends to get left by the wayside: if steganography is, or eventually becomes, the preferred tool of terrorists, can we ever thwart it? According to many experts, the answer is probably no.

The USA Today article was the first to put a number on how many stego-messages were on eBay — a number so high that many doubted it immediately. Kelley’s was also the first story to suggest that the government is specifically watching eBay, as opposed to other public Web sites. The detail that the messages “have been sent from Internet cafes in Pakistan and public libraries throughout the world” suggested that the messages found inside the image files had been encrypted, and the only thing the government was able to determine about them was the IP address of their servers.

The story had Internet libertarians crying foul. Technology reporter Declan McCullagh’s Politech mailing list, one of the last bastions of circa-1995 government wariness on the Net, featured dozens of messages from readers who were sure the piece was bogus. Politech even challenged readers to find and decode an al-Qaida missive hidden in an image file on the Web.

Libertarian skepticism does not appear to be misplaced; there are several reasons to question USA Today’s story. Kevin Pursglove, an eBay spokesman, says that while it’s possible that the company somehow missed Jack Kelley’s phone call, Pursglove and his associates in P.R. don’t recall hearing from the reporter. Moreover, eBay has never been contacted by any government agency regarding possible terrorist communications on its site. “I’m not saying what he’s reporting is not true,” Pursglove said, “but it’s just that nobody from the federal government has contacted us. We’ve got an investigations team here that has extensive contacts with federal authorities, with the FBI, the State Department, the CIA, the military. We have not had any contact at all about this.”

Salon called several federal agencies to see whether they were indeed watching eBay, but the calls went unanswered. Jack Kelley, too, did not return calls. But many security experts, even those who believe that terrorists use steganography, disputed the specifics of Kelley’s report.

Chet Hosmer, the president of WetStone Technologies, the company that first reported the possibility of hidden messages on eBay and which makes what many people say is the most advanced publicly available steganographic-detection software, said that in his research, very few messages on eBay show signs of being infected by terrorists. About one in 100,000 pictures “appears suspicious,” but a much smaller number — “one in every 15 to 20 million files” — is “something that we really believe is a real hidden message.”

Under this standard, for the government to have found 100 stego files, it would have had to have analyzed something on the order of 1 or 2 billion images. According to eBay’s first quarter financial results, the site hosted a record 138 million auctions last quarter. Extrapolating that number out for the 300 or so days since Sept. 11, we see that there have been less than half a billion eBay listings since the attacks — simply not enough to account for “hundreds” of hidden messages.

Now, this back-of-the-envelope calculation rests on several assumptions; the most important is that the government isn’t using a stego-detector more sophisticated than WetStone’s. WetStone has received funding from the Department of Defense, but Hosmer says that the government could have much fancier technology, and so it could find stego-messages at rates much higher than one in 15 million. There’s also a chance that the feds have information that allows them to narrow their search to specific sections of eBay, which would make their job considerably easier.

There’s no question that tools to hide messages in image files are easily available on the Web, and most of them are point-and-click simple to use. But as these tools scramble the message into different parts of the image file, they add some discernible “pattern” of bits — detecting stego is all about finding that anomalous statistical pattern in the code of what looks like an otherwise normal image.

Unfortunately, that process turns out to be what’s known, in the jargon, as “computationally expensive.” It’s also somewhat buggy; there’s a high false-positive rate. Consequently, when an image is suspected to have some hidden info inside it, it could take as much as 30 seconds, Hosmer said, to fully test it. That’s why you wouldn’t want to monitor all of eBay, as it would take quite some time to go through just one day’s worth of images. “With our computer power, what we tend to look at is images that we may have sources saying are suspicious, and then test those. We would act like detectives in the real world,” he said.

Acting like a real-world detective requires thinking like a terrorist, and asking yourself hard questions: If you were a terrorist, where on eBay would you hide your loot? To describe the difficulty of the task, Hosmer once coined a phrase that is often repeated by others who study steganography: “It’s not like finding a needle in a haystack. It’s like finding the right piece of straw in a haystack.”

But the task is in fact more difficult than that, because after you find what you think is your piece of straw, there’s really no way to know that you’ve got the right one. Earlier this year, Niels Provos, a graduate student at the University of Michigan, reported that after checking 2 million eBay listings, he’d found no suspect images. But when he described the study, he added, darkly, that “I can’t answer the question of whether or not there is hidden content on the Internet. My negative result doesn’t indicate that the hidden communications aren’t there.”

More recently, in response to the Politech challenge, Brian Ristuccia, a computer science student in Massachusetts, reported that he’d run some tests on Azzam.com, a pro-jihad site, and found that it had a very high positive rate for stego-images. Because these could be false positives, he’s trying to use a brute-force “dictionary attack” to break into the messages — but he doesn’t hold out hopes that he’ll find anything of substance. If he manages to crack open an image and find a message inside, Ristuccia says he’s sure the message will be encrypted. Would that mean he’s found the right straw in the haystack, the straw that hints at future terror? Short of cracking the encryption scheme — a tremendously computationally expensive task — he’ll never know.

While the challenges in fingering steganography may cast some suspicion over the USA Today report, they also don’t help make a case for the libertarian argument that the technology is relatively harmless. Neil Johnson, a steganography expert, says that he’s aware that stego could be harmful, but he says much good can come of it, too. There are many scenarios “where the observation that you and I are communicating could cause a problem for one or both of us,” he said, suggesting dictatorial regimes, military missions, that kind of thing. The argument has the flavor of a gun-rights rant — secret messages can be used for evil, but if everyone used them, society would, on balance, be better. Steganography doesn’t kill people, terrorists do.

For now, that argument doesn’t seem especially crazy; but if, after the next terrorist attack, it’s shown that the attackers used steganography to communicate with each other, governments are probably going to move against the technology.

To prevent disaster, Hosmer says that commercial sites and ISPs should take it upon themselves, now, to scrub their sites free of steganography. He suggests that sites that accept public images for posting scan each new image. He admitted that “there’s no question that that certainly benefits us, but really there is no other way to police this. There’s no way you can scan all the current information for the presence of this. It’s too vast to police it any way, but these companies could detect it early and come up with information before it’s too late.”

EBay has no plans to do this, Pursglove said. “It would have such a negative impact on the site as a whole,” he said, explaining that eBay doesn’t host its own images, which would make such scans technically difficult. EBay already has many safeguards, including requiring sellers to provide a credit card and a physical address, which would leave a paper trail to any would-be terrorist. And, Pursglove added, if the government came to eBay and told the company about some suspicious material, “We would certainly cooperate with the authorities.”

Perens’ observation held true in June, when Microsoft announced that it plans to use public-key cryptography and special cryptographic microprocessors to make the Windows operating system more secure. The initiative, called Palladium, after the mythological statue that defended ancient Athens against invaders, sits on a set of technologies that have long been in use. Neither public-key cryptography, which is decades old, nor the idea of using special hardware to bolster cryptography is new.

But it was only when Microsoft unveiled Palladium and disclosed that both Intel and AMD were willing to build hardware to support the plan that people became seriously worried about the idea of ubiquitous, cryptographically enabled and, in this case, monopolistically abetted “trusted computing.”

Many longtime critics of Microsoft have been quick to dismiss the idea as smoke and mirrors, mere public relations. It’s too complicated to ever actually work, some say, and it’s just an attempt to convince people that Microsoft is sincere about wanting to make computers “safer.” It’s also easy to see Palladium as no more than a gesture made to appease Hollywood and the recording industry, which have long been clamoring for such systems.

But what if it’s not all puffery? What if Microsoft does manage to build a foolproof Palladium and deploy it to 100 million users? Those questions elicit the really troubling scenarios. “If Microsoft has its way, there just won’t be any open-source software,” says Perens, referring to the thriving ecology of software development in which users freely share code and constantly modify each other’s applications. Perens is convinced that Palladium will let Microsoft decide which applications can run on a machine and which are simply too unsafe for public consumption — such as programs written by open-source hackers. Perens even thinks that’s the point of Palladium: “It’s designed to kill off open-source development.”

Open-source hackers aren’t the only ones who are worried. Palladium could also significantly strengthen digital rights management (DRM) — the ability of media companies to manage the content you play on your machine. At least in concept, critics say, Palladium could prevent the unauthorized copying of media of any kind, not only shutting off the MP3 file-sharing free-for-all but also interfering with the rights of consumers to make personal copies of music or movies that they purchased legitimately.

Could Palladium function as a kind of technological straitjacket, a Redmond-operated remote control over your data and, in consequence, your life? According to those who’ve looked closely at the proposal, the answer is a definite, unhelpful “maybe.” But the better question is this: Why would Microsoft want to build such a restrictive system?

“It would be a very expensive proposition just to satisfy Hollywood,” says David Farber, the chief technologist of the Federal Communications Commission. Microsoft itself says that Palladium is not meant as a vehicle for DRM — that it will play anything users want it to play, whether that’s an MP3 grabbed from KaZaA or an illegally copied “Simpsons” episode. More to the point, if Microsoft did come up with a restrictive hardware and software solution that clamped down on user freedom, people would just find a way to work around it, say some observers. Either folks will break the system, which is not inconceivable, or they’ll use another system. And from what we know about Bill Gates, this much is clear: The thing that keeps him up at night is the thought of people using other systems.

Microsoft is going out of its way to tell users that Palladium won’t stop them from doing what they like on the Internet and that, scout’s honor, everything the system does will be to the good. According to Peter Biddle, a Microsoft product manager, Palladium is nothing more than an elegant solution to the vexing problem of keeping people secure on the Internet, a goal that Gates has set as one of the company’s main objectives.

The strongest part of Palladium will be its ability to determine whether a given software application should run on a machine. The system will be shipped with these functions turned off, but “we actually think it’s likely that users will say, ‘I’m only going to run code that’s been signed,’” Biddle says. By “signed,” Biddle means that the application has been cryptographically tagged by a “signing authority.” The Palladium system would run the code only if the user has approved that specific authority.

Theoretically, this would make computing much safer. If you set your machine to run code that’s been signed, then the many errant applications — viruses, spyware, adware and the like — that float into your machine without your express knowledge would find no shelter on your desktop. “For years we’ve dealt with computer systems that were basically not secure at all,” the FCC’s Farber says. “This could be a step in the right direction.”

Farber is no Microsoft stooge. He testified against the company during its antitrust trial, telling the court that Internet Explorer was not actually an integral part of the operating system, as Microsoft had claimed. But he appreciates that Microsoft is taking the initiative in security, because businesses and, to a lesser extent, home users are afraid that computer systems don’t keep their data safe, and because the situation is becoming worse.

Farber concedes, though, that whether or not one thinks of Palladium’s architecture as a boon to security “depends on what you believe Microsoft’s long-term aims are. If you believe it’s to stimulate commerce and stimulate security, it’s a step in the right direction.” But if you’re more “neurotic” than that, Farber says, and if you’re perhaps given to suspicions that Microsoft always makes decisions with the aim of frustrating competitors of the Windows empire rather than for the good of consumers, you might have a different view of the same architecture.

“Until we see it, until we actually look at the code, until we go through the whole process and see how the whole system will work, we won’t know what it’s like,” Farber says. “If they do it all right, it might work — but it can be misused.”

A key question about Palladium’s process is this: Who will be authorized to sign code? Microsoft says that it will have no say in that process. There will be multiple code-signing authorities, and they will be “self-vetted,” Biddle says. “They go into business saying, We’re here to sign this code.” Microsoft would make no claim about the safety of that code, Biddle says, and the code signers would not be compelled to do so either. A cryptographic sign, then, would essentially work like a Good Housekeeping seal of approval, and you could decide whether to trust the code based on what you think of the signer. In theory, an organization like the hacker/cracker publication 2600 could sign code, Biddle says, as could open-source companies and free software advocates or whoever else people trust.

“We actively encourage and are pursuing a strategy that says the hardware runs everything it runs today,” Biddle says, implying that Palladium is not designed specifically to prevent certain kinds of software from being used.

But Perens and others note that the final authority to give signing status to a certain group or individual — or to revoke signing status — would necessarily lie in a central location, presumably with Microsoft. Could Microsoft abuse that status? Would the company charge to license its keys, or give preferential treatment to some companies? And even if Microsoft plays fair with the keys, Perens says, and if it’s just as easy for scrappy developers to sign their code as it is for corporations, the whole system would still be impractical for open-source developers.

By definition, the code in an open-source application is not set in stone. The whole point of the General Public License, the license under which Linux-based operating systems are offered, is to allow people to modify code ad infinitum. But under Palladium, an application that has been modified loses its signature. Each new version of an application, therefore, would presumably need a signature before it could run on a system.

Perens says that “what is new here is that the customer’s PC is getting hardware with the specific purpose of constraining the customer. Never before has a customer received a speed governor on his car — and this is worse than a speed governor. It’s like saying, ‘You may never drive into this part of town.’”

It’s worth pausing to think about Perens’ example for just a second. Surely some lawyer somewhere has suggested to one of the Big Three automakers that adding speed governors to its fleet could save the company a penny or two in legal costs. So why don’t we have speed governors in our Fords? Right — because you wouldn’t buy a car that’s constrained by a speed governor, just as you wouldn’t want to buy a CD that doesn’t play in your DiscMan, or pay for a music subscription service that doesn’t allow you to permanently record its music.

Brian Behlendorf, the co-founder of the open-source Apache Web Server Project, calls Palladium “subtractive” for just that reason. “It feels very much like a genie is out of the bottle on this,” he says. “Let’s say they ultimately get this fierce line of control around this. It could make people look at hardware alternatives, because it’s a subtractive value to end users. Look at the interest in region-free DVD players, or mod chips for the Xbox.”

In other words, people will look for other options if Microsoft gets too restrictive. And that’s the last thing Microsoft wants.

Microsoft argues that Palladium can always be switched off by users who think it’s bad news. If Palladium becomes ubiquitous, critics respond, that may not be an option.

“If you turn it off, then you are an island,” says Perens. “You can’t communicate with others. Everyone will be using this DRM, and you can’t view Web pages.”

Just about everyone who commented on Palladium feared the possibility of being compelled to use the system because someday it will be the only platform that will play content. “That’s the core of the Palladium thing,” says Miguel de Icaza, founder of the open-source software company Ximian. “What they want is to have the media companies feel safe.”

The media companies, de Icaza’s argument goes, would design their CDs, DVDs, e-books, Web pages and all their other content to show up only on a Palladium system. Media that’s been tampered with — a ripped CD track converted to an MP3 — would lack a cryptographic watermark, and Palladium would therefore refuse to open it.

Media companies yearn for such safety. Earlier this year, they got Sen. Ernest “Fritz” Hollings, the South Carolina Democrat who chairs the Senate Commerce Committee, to introduce a bill that could provide them with some measure of legislative safety. Hollings’ bill, the Consumer Broadband and Digital Television Promotion Act (CBDTPA), would require virtually all electronic devices, from CD players to PCs, to include copy-protection mechanisms. And now, say critics of Palladium, Microsoft is offering a computer system designed to satisfy the Hollings bill, a system that has at its heart what Ross Anderson, a Cambridge computer scientist, calls the “Fritz chip.”

Microsoft denies that Palladium is a Trojan horse that will allow it to slip DRM into computer systems. “Turning Palladium on is not the same as turning DRM on,” says Biddle. “This will allow you to do a much better job of protecting your privacy. So from an end-user benefit perspective, that’s what we’re focusing on. We’re not focusing on creating a DRM infrastructure”

He acknowledges that Palladium will strengthen DRM, and it’s conceivable, he says, that you may need to turn on DRM to play some content. But “that’s up to the seller of the DVD, and it’s also our belief that any content worth stealing is going to wind up on KaZaA or whatever, and we can’t stop that. DRM and Palladium don’t do anything to prevent people from downloading movies off the Web. We can’t tell the difference between a home movie and a DVD. To Palladium, they are the same. There are some people who believe that with watermarking you could tell the difference, but we aren’t believers that watermarks are robust enough to do that, so we’re going to play it all.” Even your cherished collection of stolen MP3s, Biddle says, would work without any problems on Palladium.

Biddle’s comments are noteworthy, as they appear to run contrary to overall Microsoft policy. The company offers its own DRM programming suites, and it has always publicly supported Hollywood. Through the Business Software Alliance, it has also fiercely protected its own intellectual property.

It’s also hard to tell whether Biddle really means it when he says that Palladium won’t be able to stop media theft — but he’s right. As proved by Ed Felten, the Princeton computer science professor who cracked SDMI, the recording industry’s ballyhooed DRM technology, watermarking doesn’t ensure media security. Invariably, one whiz-kid hacker or Ivy League coder will find a way to get around such a system, and short of throwing people in jail (not out of the question), DRM will break. Microsoft, perhaps, is wise to this.

Biddle’s arguments about DRM are also somewhat muddied by the fact that late last year Microsoft was quietly granted a patent for just what he says it’s not building: “The Digital Rights Management Operating System,” protected by U.S. patent numbers 6,330,670 and 6,327,652.

“In a very real sense, the legitimate user of a computer can be an adversary of the data or content provider,” one of the patents says. “‘Digital rights management’ is therefore fast becoming a central requirement if online commerce is to continue its rapid growth … If measures are not taken, traditional content providers may be put out of business by widespread theft, or, more likely, will refuse altogether to deliver content online.”

Chris Hoofnagle, of the Electronic Privacy Information Center, says that such a system would harm, not secure, your online privacy. “Many of the DRM systems rely on personal information, and it basically ties a person to a piece of content,” he says. “The Palladium system and many of the services offered are going to depend on your identity — but Microsoft has been very clever in presenting this as a system that can make you safer on the Internet.”

Microsoft, for its part, says that it’s very sensitive to claims that Palladium will be bad for privacy, and it has been offering to work with privacy groups to make sure it gets the system right. The Electronic Frontier Foundation, an online-rights advocacy group, has been briefed on Palladium and it says it’s studying the system and will soon offer an opinion.

What the EFF and the rest of the industry probably want to know is whether Palladium, in the end, will be good for regular people. For all its faults, Microsoft is not known for kicking its customers in the teeth. If Palladium stops viruses, doesn’t constrain your machine, and doesn’t invade privacy — above all, if people are allowed to control Palladium, rather than vice versa — would the system be so bad?