by Cory Doctorow
European Affairs Coordinator, Electronic Frontier Foundation
28 September 2005
This text is dedicated to the public domain, using a Creative Commons public domain dedication:
Copyright-Only Dedication (based on United States law)
The person or persons who have associated their work with this document (the "Dedicator") hereby dedicate the entire copyright in the work of authorship identified below (the "Work") to the public domain.
Dedicator makes this dedication for the benefit of the public at large and to the detriment of Dedicator's heirs and successors. Dedicator intends this dedication to be an overt act of relinquishment in perpetuity of all present and future rights under copyright law, whether vested or contingent, in the Work. Dedicator understands that such relinquishment of all rights includes the relinquishment of all rights to enforce (by lawsuit or otherwise) those copyrights in the Work.
Dedicator recognizes that, once placed in the public domain, the Work may be freely reproduced, distributed, transmitted, used, modified, built upon, or otherwise exploited by anyone for any purpose, commercial or non-commercial, and in any way, including by methods that have not yet been invented or conceived.
Note: this essay is derived from notes for an invited talk to HP Research on DRM. The talk was not delivered verbatim, nevertheless, this is a good feel for what I said that day. For the text of an earlier talk on this subject delivered to Microsoft Research, see http://craphound.com/msftdrm.txt.
The canonical version of this talk live at http://craphound.com/hpdrm.txt.
I work for the Electronic Frontier Foundation, a member-supported charitable organization that works to uphold the public interest in technology law, policy and standards. For nearly four years, I've spent my time attending DRM standards meetings, consortia, and treaty meetings at the United Nations. In that time, again and again, I've seen tech giants like HP take suicidal measures to voluntarily cripple their products to make them more palatable to a few entertainment companies, even though this measure makes them less palatable to virtually all of your paying customers.
Nothing epitomized this more than Carly Florina's inaugural CES address in which she promised to put DRM in every HP product. Reading that in my office in San Francisco (I live in London now), I thought, well, hell, I guess I'm not buying any more HP products. I'm pretty sure I'm not the only one.
I've had innumerable conversations with engineers, lawyers and execs about DRM, but it's rare that I get the chance to systematically explain how DRM fails as a technology, as a moral proposition, and as a commercial initiative. I'm grateful that HP has given me that chance today. I'm looking forward to your questions after my talk.
Now, onto the talk, in which I will try to address the security, moral and commercial aspects of DRM.
There is no such thing as "security" in the abstract. You can't be made "secure." You can only be made "secure" *against a specific attack*. All security discussions must begin with an analysis of a threat and a proceed to address that threat with countermeasures.
In discussions of DRM, radically different threat-models are usually conflated to sow confusion and to disguise the implausibility of DRM. In the paper at hand (as in many other cases), privacy-protection is conflated with use-restriction. But these have totally different threat-models:
In privacy scenarios, there is a sender, a receiver and an attacker. For example, you want to send your credit-card to an online store. An attacker wants to capture the number. Your security here concerns itself with protecting the integrity and secrecy of a message in transit. It makes no attempt to restrict the disposition of your credit-card number after it is received by the store.
In DRM use-restriction scenarios, there is only a sender and an attacker, *who is also the intended recipient of the message*. I transmit a song to you so that you can listen to it, but try to stop you from copying it. This requires that your terminal obey my commands, even when you want it to obey *your* commands.
Understood this way, use-restriction and privacy are antithetical. As is often the case in security, increasing the security on one axis weakens the security on another. A terminal that is capable of being remotely controlled by a third party who is adversarial to its owner is a terminal that is capable of betraying its owner's privacy in numerous ways without the owner's consent or knowledge. A terminal that can *never* be used to override its owner's wishes is by definition a terminal that is better at protecting its owner's privacy.
The threat model for DRM is that an unscrupulous user will be able to download an asset for free from the Internet instead of going through a conditional access billing gateway. Additionally, DRM seeks to give rightsholders the ability to restrict the use of assets after receipt to enforce restrictions that are not related to copyright (e.g. remote viewing, region-control).
A service operator can ensure that 100 percent of the assets behind her conditional access system are wrapped with DRM, which means that everyone who uses the system will receive media that is locked with DRM. The system fails not when the DRM is cracked, but when a user gains access to a non-DRM file, or when a user does not pay for access.
Every file that is locked with DRM inside a conditional access system is also available on the public Internet without DRM. In order for DRM to be effective, a user must first freely choose to acquire the DRM version over the non-DRM version.
The presence of DRM *cannot* entice a user to make use of the conditional access system to acquire his media. Indeed, DRM acts as a disincentive (there is no user who woke up this morning crying out for a way to do less with her music). Where users buy DRM-locked files, it is *in spite of* the DRM, or in ignorance of the DRM, but never *because* of the DRM.
A familiar refrain from rightsholders is that "you can't compete with free." It is certainly true that when your costly product is inferior (because of use-restrictions) to the free alternative, it will be hard to compete with free.
In the DRM world, security is breached so long as there is any person with the wherewithal to make a cleartext copy of an asset and put it on the Internet. In practice, this happens with amazing swiftness. Big Champagne, a company that monitors P2P networks, says that iTunes-only tracks (e.g. assets that are only released within DRM wrappers) typically appear on P2P networks less than three minutes after they are released to the iTunes Music Store.
To succeed in an attack against a DRM system, a user need not know how to break DRM, she only needs to know how to search Google or another general-purpose search tool for a copy that someone else has already rendered in the clear.
The privacy threat model generally revolves around accidental disclosure and subsequent publicity. A common example of privacy breach is an unscrupulous hospital worker who discloses the identities of HIV-positive patients.
It is suggested that an iTunes Music Store-like model could defend against this attack: a conditional access system restricts access to a health record unless a valid credential (e.g. a password or smartcard) is presented. A DRM system allows for later revocation of access once it has been granted. However, as Don Marti points out, this is poor security indeed:
"Deploy DRM and you can keep employees from forwarding embarrassing email to the media. That sounds like the answer to network-illiterate managers' prayers, but if it's juicy enough to leak, it's juicy enough to write down and retype.... Bill Gates pitch[ed] DRM using the example of an HIV test result, which is literally one bit of information. If you hired someone untrustworthy enough to leak that but unable to remember it, you don't need DRM, you need to fix your hiring process."
Don Marti, editor in chief, Linux Journal
Privacy almost always includes an element of personal/political power. Children want to be private from their parents. Employees want privacy from their bosses. Political dissidents want privacy from the Chinese secret police.
For "privacy DRM" to work, the defender needs to be in a position to dictate to the attacker the terms on which he may receive access to sensitive information. For example, the IRS is supposed to destroy your tax-records after seven years. In order for you to use DRM to accomplish the automatic deletion of your records after seven years, you need to convince the IRS to accept your tax records inside your own DRM wrapper.
But the main reason to use technology to auto-erase your tax-records from the IRS's files is that you don't trust them to honor their promise to delete the records on their own. You are already adversarial to the IRS, and you are already subject to the IRS's authority and in no position to order it to change its practices. The presence or absence of DRM can't change that essential fact.
This is a classic "who will bell the cat?" problem. Inventing new and better-functioning bells doesn't make getting them attached to the cat's collar any easier.
Many of the restrictions that DRM is used to enforce are unrelated to copyright, and no DRM system can accurately model copyright, which is highly fact-specific.
Copyright is a limited monopoly over the public copying, performance, display and adaptation of original works. Copyright governs the ability of commercial entities and a few noncommercial entities to make copies, display them, etc.
Copyright does *not* confer the right to control "remote viewing" -- the ability to store a show in one place and watch it in another. It does *not* confer the right to control timeshifting. It doesn't confer the right to control regional playback, as with DVDs that can only be viewed on a US player or a European players. Copyright does *not* confer the right to control re-sale or lending of lawfully acquired works.
Copyright is used to extend the creator's monopoly into all kinds of realms, though. Take the so-called "Authorized Domain", a trendy DRM concept that confers on rightsholders the right to define valid familial arrangements, something so far remote from copyright as to be in an entirely different universe. In venues where the Authorized Domain is being planned, designers are torn between two different potential implementation models, both of which are totally unacceptable:
Only so many devices may join the domain (as with Apple's five-device authorization limit for iTunes). This has many unacceptable failure modes, including the inability to deactivate lost, stolen or damaged devices, as well as arbitrary limits on family size.
In this model, a series of tests are applied, including tests for proximity, tests for existing domain size, strategies for re-accumulating domain credits, and proprietary tests. These tests are logically represented on flowcharts that no end-user or retailer can possibly understand (especially given the presence of proprietary tests). Any customer who asks a retailer, "Will this device be able to join my domain?" will inevitably get the answer: "maybe."
Most unacceptable is the presence of "corner cases" like divorced families with joint custody arrangements among several children, whose devices may be restricted from belonging to more than one domain, or blended households created in extremis (your father being sent to an old folks' home, your daughter moving into a student house), that are surely households, even if they are not traditional families, and that may fail the tests on domain size.
DRM is often characterized as the outcome of a negotiation: "You may have access to my song if you accept my restrictions." But DRM always gives rightsholders the ability to unilaterally renegotiate the terms of the deal to take away rights you acquired when you got your device and media.
For example, many updates to iTunes contain new restrictions on the music you purchase. In the past 18 months, iTunes has instituted the following new restrictions:
You buy a song on day one and can do ten things with it. A few weeks later, you can only do nine things with it. Then eight. Then seven.
Last week, many TiVo owners discovered that several of the free-to-air and cable shows they received with their PVRs could not be saved indefinitely, and would be automatically deleted after a set period.
Last year, Comcast PVR owners discovered that all their stored episodes of Six Feet Under were deleted a few weeks before the DVD came out.
The right to store your music and movies, the right to watch your movies in any country you find yourself in, the right to timeshift and space-shift, the right to re-sell, the right to loan, the right to share your media with your family regardless of your familial arrangements -- these rights all belong to the public. Copyright law reserves these rights from control by rightsholders.
DRM is a mechanism for unbalancing copyright, for betraying the statutory limitations on copyright, for undermining the law itself. By granting rightsholders the ability to unilaterally confiscate public rights under copyright, DRM takes value out of the public's pocket and delivers it to rightsholders.
When you acquire a car, you acquire the right to charge your phone off its cigarette lighter. No car owner has to assign that right to you. Even if the car manufacturer thinks it can make big bucks by selling the exclusive right to charge phones in its car to Nokia, nothing prevents you from charging your Motorola phone from the lighter.
More complex are the rights reserved to the public under the banner of fair use. Fair use is the copyright doctrine that allows users to make uses *even if the rightsholder objects*. For example, critics, parodists, educators, archivists and disabled people all have certain rights to use copyrighted works without the permission from rightsholders. In order for a DRM system to permit you to extract some video for the purposes of making a parody, but stop you from doing this for the purposes of burning the movie to a CD and selling it on eBay, the DRM system has to be capable of reading your mind and determining why you want to make your use.
The gradual tightening of DRM screws will alienate ever-larger groups of customers. There are some who believe that if you turn the heat up gradually enough, the customer will never notice that she has been boiled. History suggests otherwise. The repeated disastrous attempts to introduce DRMed CDs into the marketplace tells us once a customer is accustomed to a use, she is unlikely to accept a product that restricts it.
HP is under no obligation to play by the entertainment industry's rules in order to gain access to content. Format-shifting, time-shifting and space-shifting are legal practices with long and honorable traditions (indeed, Apple's own iTunes software contains a mechanism to format- and space-shift your CDs by ripping them to MP3, as does Microsoft's Media Player).
However, when tech companies seek a closer relationship with the entrainment industry, they find themselves in the position of having to offer means for restricting the use of their products in ways that the market generally rejects -- no end-user buys products because of their DRM.
The worst-case scenario is to end up in a situation like the Blu-Ray/DVD-HD wars. The two consortia responsible for these competing formats are competing to please the entertainment industry by adding more and more onerous restrictions to their technologies, restrictions that raise the manufacturing costs while reducing the commercial viability of their products.
HP need not follow this disastrous strategy. Practically every device in the field has one or more analog outputs. It is both possible and legal to connect digital recording devices to these outputs and make legal near-perfect digital copies that can be played back and manipulated on devices without Hollywood's blessing. Devices such as the Slingbox, the Orb, and Mythtv all do this today.
These devices play perfectly to the core strengths of the tech and telecoms industry. PC vendors who provide flexible set-top boxes that ease the pain of recording and librarying AV material will create markets for ever-more-capable set-top boxes that have larger and larger storage capacities, as well as backup solutions, service and troubleshooting, etc.
Current models for trusted computing conflate many features that are useful to the user with many that undermine user privacy, investment in content, and data-integrity.
On the positive side, trusted computing allows for superior countermeasures against spyware and other malicious software. It contains crypto accelerators that safeguard communications integrity and secrecy. It eases the pain of managing end-to-end crypto for private communications.
On the negative side, trusted computing can enforce policies against a user's wishes. Trusted computing can be used to block the use of interoperable products (e.g., to force a user to use Internet Explorer instead of Mozilla by allowing remote parties to reliably distinguish among the two), and to block or complicate the backing up or migration of user data. Additionally, trusted computing can be used as a superior enforcement mechanism for DRM restrictions, particularly those that seek to unilaterally renegotiate the terms under which content is acquired.
This need not be. "Owner override" is a conceptual model for modifying trusted computing hardware to retain all of its user benefits while eliminating the dangers posed by allowing a device to enforce policy against its owner's wishes.
For more information on "owner override" please see Electronic Frontier Foundation Staff Technologist Seth Schoen's excellent paper on the subject:
Owner Override works by empowering a computer owner, when physically present at the computer in question, deliberately to choose to generate an attestation which does not reflect the actual state of the software environment -- to present the picture of her choice of her computer's operating system, application software or drivers. Since such an attestation can only be generated by the computer owner's conscious choice, the value of attestation for detecting unauthorized changes is preserved. But the PC owner has regained fine-grained control, even in a network environment, and the PC can no longer be expected to enforce policies against its owner. Owner Override removes the toolbox that allows the trusted computing architecture to be abused for anti-interoperability and anti-competitive purposes. It restores the important ability to reverse engineer computer programs to promote interoperability between them. Broadly, it fixes trusted computing so that it protects the computer owner and authorized users against attacks, without limiting the computer owner's authority to decide precisely which policies should be enforced. It does so without undermining any benefit claimed for the TCG architecture or showcased in Microsoft's public NGSCB demonstration. And it is consistent with TCG's and most vendors' statements about the goals of trusted computing.
I can hardly fault HP for embracing the received wisdom on DRM. However, the received wisdom is rarely a path to commercial success. In the global marketplace, HP has numerous competitors, from giants to smaller, nimbler firms -- and if any company has an appreciation of the potential of two guys in a garage, it should be this one.
The question isn't *whether* one of these companies will defect from the DRM game, but *when*. The first to market with better, more powerful, more capable devices will emerge the clear winner.
I don't believe HP can afford to sit tight and hope that the unspoken agreement not to anger Hollywood will hold.
HTML-ed by Branko Collin on October 1, 2005, based on the version at http://craphound.com/hpdrm.txt from the same day. I don't claim copyrights on my mark-up, and I won't sue for copying, even if the law of your land grants copyrights for mark-up to me anyhow.
Apart from the mark-up and these notes, my only changes were the addition of the word "by" in the byline, and the conversion of an English style publication date to an international one.