Sunday 25 February 2018

Peaceful coexistence, jurisdiction and the internet

I am typing this on the transatlantic flight to Canada, destination the Global Internet and Jurisdiction Conference in Ottawa (#OttawaGIJC).  This seems like a good time to set down some propositions that, even if they do not command universal agreement, I believe are central to the debate about internet and jurisdiction.

First, something about what we mean by jurisdiction. We shouldn’t get too hung up on this, because jurisdiction is mostly used just as shorthand for cross border legal liability. But lawyers use the single word jurisdiction to mean several different things and, if we are not careful, we can end up talking at cross-purposes.

In a nutshell, there is:
  • Prescriptive jurisdiction. This is simply the assertion, however theoretical, as to the reach of a local law. If the UK had passed a law making it illegal for UK citizens to read the banned book Spycatcher when visiting the USA, that would be an assertion of prescriptive jurisdiction. Another example is when a court makes an order requiring something to be done (or not done) in a foreign country.
  • Adjudicative jurisdiction. This is when a court determines that it has the ability (or not) to hear and decide a case. It also describes a court determining which country’s law it should apply to the case before it. In civil litigation between private parties the court may end up applying its own law or a foreign law. In a criminal prosecution the court will either apply its own country’s law or refuse jurisdiction.
  • Enforcement jurisdiction. This engages territoriality issues most strongly, since taking enforcement measures in another country is rather like sending troops across the border – an invasion of territorial sovereignty unless done with the consent of the other state. 
  • Investigatory jurisdiction. Sometimes regarded as a subset of enforcement jurisdiction, investigatory jurisdiction has become such an important topic for the internet (think of the Microsoft Warrant case pending before the US Supreme Court, or the Yahoo case that went up to the Belgian Supreme Court and back three times) that it deserves its place as a separate category.The issue here is when and how it is legitimate for law enforcement or intelligence agencies, with or without court orders, to demand that communications (or their associated data) held overseas be handed over.  As with enforcement jurisdiction this involves direct assertion of state power within the territory of another country – either where the data is held, where a foreign company is established, or both.

It gets more complicated than that. International law has developed principles around when it is legitimate for a state to act extraterritorially.  They form the background to a discussion of how jurisdiction should apply to the internet. But we should not necessarily be hamstrung by the existing order of things when debating the question of what rules should look like in the internet age.

Yes, the internet is different. Time was when people would say that the internet was just another new medium. We have coped with cross-border issues as a result of international communications and satellite broadcasting, so why do we need new rules for the internet? 

The internet is not wholly different, but some of its features are sufficiently different to demand reconsideration of the old rules. 

Individuals by the million are authors and publishers as well as readers. Their content is instantly accessible worldwide. Conversely, the effect of a national law or court injunction is amplified by the internet. An order can have effects in other countries that the same instrument would not have in the offline world. Cloud computing means that data is volatile. It may not stay in the same country from one second to the next, and fragments of one item of content may be split between data centres in different countries. All these things render the internet not only different, but materially so.

It’s not about whose law is better. If you arrive at a jurisdiction conference determined to demonstrate the superiority of your own local, national or regional law over that of every other country, then you are at the wrong conference. Jurisdiction rules are about resolving friction between different legal systems in as agnostic a way as possible, not about ensuring that the best (in someone’s view) law wins.

It’s not about global harmonisation. Perhaps you harbour an ambition of achieving global consensus about the substantive laws that should apply to the internet. That may be a noble goal (albeit there is also merit in diversity of laws) but it is a different project.  Jurisdiction rules presuppose different laws in different countries, albeit admittedly it is easier to reach agreement on jurisdictional rules when the underlying laws are more closely aligned. Nevertheless, while a jurisdiction project can aim to create international norms at the level of metalaw (rules about rules), creating uniform substantive law is not its goal.

Comity is not enough. Resolving jurisdictional frictions is often seen through the prism of comity. Comity has two aspects: the need as far as possible to recognise a legitimate assertion of state power by foreign countries, even if that may have some cross-border spillover effects; and conversely, the need to avoid treading on the toes of other foreign states when making orders that may have effects in other countries (but in both cases bearing in mind that spillover effects are likely to be greater on the internet than offline).

However, comity is a state-centric concept. It treats states as proxies for the rights and interests of their citizens. Extraterritorial legislation and court orders not only engage the sensitivities of other states, but directly affect individuals in other countries, engaging their universally recognised fundamental rights of, for instance, privacy or freedom of expression. 

Those individuals’ interests stand to be considered separately from the question of whether the sensitivities of another state are likely to be engaged by a particular legal instrument. Failure to engage in separate consideration can lead to the kind of reasoning adopted in the Equustek case, where the Supreme Court of Canada concluded that since protection of intellectual property was the kind of interest that another state would be expected to protect, its sensitivities would not be engaged and there was no issue of comity.

The SCC did not go on to ask whether, and if how the freedom of expression interests (the right to receive and seek out information) of citizens of other countries might be engaged by the particular assertion of rights in that case. That is of particular relevance in intellectual property cases since intellectual property rights are themselves generally territorial, so that a person may own rights only in some countries and not others; or may own rights of different scope in different countries.

We need brakes as well as accelerators. The jurisdictional problems of the internet manifest themselves in both underreach and overreach. There are situations where arrangements between states are no longer providing adequate means to obtain evidence to support criminal investigations. We can no longer assume that the evidence relating to a domestic crime will be held domestically. It could as easily be in a data centre abroad.  That would suggest a need to improve procedures for obtaining cross-border evidence.

Conversely, we have situations in which domestic legislatures, agencies and courts are at risk of overreaching in the cause of giving maximum effect to their local laws. That can result in the de facto imposition of those laws in countries with different laws. The concern here is the need for jurisdictional self-restraint.

The challenge is to forge rules that enable cross-border reach when appropriate, yet prevent the exercise of jurisdiction when not appropriate. The same kinds of rules are unlikely to achieve both. An approach that enables a court to weigh up and balance a series of factors in deciding whether or not to make an extraterritorial order may have desirable flexibility for the first case. But where risk of jurisdictional overreach is concerned a multi-factorial approach may be more enabling than restraining. Hard stops are likely to be required.

Peaceful co-existence requires compromise. The premise of jurisdiction rules is that nation states have different laws.  The objective where the internet is concerned should be to achieve peaceful co-existence between conflicting national regimes while protecting to the greatest possible extent universal values such as freedom of expression and privacy.

Peaceful co-existence cannot be achieved without compromise. That means taking a broader view than simply a laser-like focus on securing the effectiveness of one country or region’s most cherished laws. It may mean accepting that your country’s citizens can, if they try hard enough, find somewhere on the internet content that complies with another country’s laws and not your own.

(For more on this final topic see Cyberborders and the Right to Travel in Cyberspace, my chapter in The Net and the Nation State (2017 CUP, ed Uta Kohl).)

Thursday 22 February 2018

Illuminating the Investigatory Powers Act

As full implementation of the Investigatory Powers Act (IPAct) draws closer we can usefully ponder some of its more ticklish points of interpretation. These will serve to delineate the IPAct's powers, crystallise the legislation's procedural requirements and determine who can be compelled to do what.

Unlike its predecessor, the Regulation of Investigatory Powers Act 2000 (RIPA), the IPAct comes with expectations of openness and transparency.  The Act itself exposes a panoply of powers to the public gaze.  But despite its 300 pages of detail, decisions will still have to be made about the meaning of some provisions and how they are to be applied.

Previously such legal interpretations have tended to come to light, if at all, as a consequence of the Snowden revelations or during litigation brought by civil liberties organisations. Examples include the meaning of ‘external’ communications under RIPA, the legal basis for thematic interception warrants under RIPA, and the use of S.94 Telecommunications Act 1984 powers to acquire bulk communications data from telecommunications companies.

In the field of surveillance, hidden legal interpretations influencing how powers are wielded are in substance as much part of the law as the statute that grants the powers.  This can be problematic when a cornerstone of the rule of law is that laws should be publicly promulgated. People should be able to know in advance the kind of circumstances in which the powers are liable to be used and understand the manner of their exercise. According to jurisprudential taste, secret law is either bad law or not law at all.

The new Investigatory Powers Commissioner has an opportunity to bring to public view legal interpretations that will mould the use of the IPAct's surveillance powers. 

Most IPAct powers require approval by a Judicial Commissioner or, as now proposed for communications data acquisition, a new Office for Communications Data Authorisations. The Judicial Commissioner or other reviewer may have to form a view about some provision of the Act when approving a warrant or notice.  Some interpretations may have significance that goes wider than a single approval.

Under the IPAct there is scope for an adopted interpretation to be published if that can be done without breaching the Commissioner's responsibilities not to act contrary to the public interest, nor prejudice national security or the prevention or detection of serious crime or the economic well-being of the UK.

What interpretations of the IPAct will have to be considered? The most heavily debated has been the level of scrutiny that Judicial Commissioners are required to apply to Ministerial decisions to issue warrants and technical capability notices. Gratefully donning my techlaw hat, I shall leave that problem to the public and administrative law experts who have been mulling over it since the draft Bill was published in November 2015.

Approval decisions will typically involve assessments of necessity and proportionality. These will by their nature be fact-sensitive and so more difficult to make public without revealing operational matters that ought to remain secret. Nevertheless some general approaches may be capable of being made public.

Among the most likely candidates for publication will be points of statutory construction: aspects of the IPAct's language that require a view to be taken of their correct interpretation.  

I have drawn up a list of provisions that present interpretative challenges of varying degrees of significance. Some of the points are old hobbyhorses, dating back to my comments on the original draft Bill. Others are new. No doubt more will emerge as the IPAct is put into practice.


Selection for examination

What is the issue?

Under a bulk interception warrant what kinds of activities count as selection for examination of intercepted content or secondary data? While the question can be simply put, the answer is not so easy.

Why is it significant?

Selection for examination underpins three provisions of the IPAct.

First, a separate targeted examination warrant must be obtained before selecting intercepted content for examination by use of criteria (such as an e-mail address) referable to an individual known to be in the British Islands, if the purpose is to identify the content of communications sent by or intended for that individual. (S.152(4)) (However, a targeted examination warrant is not required for secondary data. As to what is meant by secondary data, see below.)

Second, it is an offence (subject to applicable knowledge and intent thresholds) to select intercepted content or secondary data for examination in breach of the Act's safeguards. (S.155)

Third, a bulk interception warrant authorising selection for examination must describe the manner in which intercepted content or secondary data will be selected for examination and the conduct by which that activity will be secured (S.136(4)(c)).

The S.136(4)(c) requirement is new compared with the equivalent provisions of RIPA. Curiously, it is not referred to in the draft Interception Code of Practice

It is important to know what activities amount to selection for examination.  This is a particular issue with automated processing.

Possible interpretations?

Examination means being read, looked at or listened to (S.263) But what activities are caught by selection for examination? How close a nexus does there have to be between the selection and any subsequent examination?  Does there have to be a specific intention to examine the selected item (for instance when an analyst makes a search request on a database)? Does selection for possible examination suffice?  (It is perhaps of interest that David Anderson Q.C.'s Bulk Powers Review at para 2.17 discusses under the heading of ‘Selection for Examination’ the use of strong and weak selectors to select material for “possible examination” by analysts.)

The Draft Interception Code of Practice describes a sequence of steps from obtaining the data through to examination by an analyst. It uses the term 'selection for examination' in ways that could refer to both selection by the analyst and intermediate processing steps:
"In practice, several different processing systems may be used to effect the interception and/or the obtaining of secondary data, and the selection for examination of the data so obtained. 
These processing systems process data from the communications links or signals that the intercepting authority has chosen to intercept. A degree of filtering is then applied to the traffic on those links and signals, designed to select types of communications of potential intelligence value whilst discarding those least likely to be of intelligence value. As a result of this filtering, which will vary between processing systems, a significant proportion of the communications on these links and signals will be automatically discarded. Further complex searches may then take place to draw out further communications most likely to be of greatest intelligence value, which relate to the agency’s statutory functions. These communications may then be selected for examination for one or more of the operational purposes specified in the warrant where the conditions of necessity and proportionality are met. Only items which have not been filtered out can potentially be selected for examination by authorised persons." (emphasis added)
If selection for examination encompasses only the action of an analyst querying a database then S.136(4)(c) would still require the warrant to describe the manner in which an analyst could select content or secondary data for examination. That could include describing how analysts can go about searching databases. It might also cover the operation of Query Focused Datasets (databases in which the data is organised so as to optimise particular kinds of queries by analysts).

But does selection for examination exclude all the automated processing that takes place between bulk capture and storage? There appears to be no reason in principle why automated selection should be excluded, if the selection is 'for examination'.  

Details of the kinds of automated processing applied between capture and storage are mainly kept secret.  However some clues beyond the draft Code of Practice can be obtained from the Intelligence and Security Committee Report of March 2015 and from the Bulk Powers Review.  The Bulk Powers Review describes a process that uses ‘strong selectors’ (telephone number or email address) to select items in near real time as they are intercepted:

“As the internet traffic flows along those chosen bearers, the system compares the communications against a list of strong selectors in near real-time. Any communications which match the selectors are automatically collected and all other communications are automatically discarded.”

Such selection against a list of e-mail addresses or telephone numbers of interest is not made for any purpose other than examination, or at least possible examination. But does it count as selection for examination if (as described in the Bulk Powers Review) a further triage process may be applied?

“Even where communications are known to relate to specific targets, GCHQ does not have the resources to examine them all. Analysts use their experience and judgement to decide which of the results returned by their queries are most likely to be of intelligence value and will examine only these.”

Weaker selectors may relate to subject-matter and be combined to create complex non-real time queries which determine what material is retained for possible examination after triage. Pattern matching algorithms could perhaps be used to flag up persons exhibiting suspicious behavioural traits as candidates for further investigation.

The question of which, if any, of these processes amount to selection for examination is of considerable significance to the operation of the processes mandated by the IPAct.

Secondary data

What is the issue?

'Secondary data' under the IP Act has been extended, compared with RIPA's equivalent ‘related communications data’, so as to include some elements of the content of a communication. However the definition is difficult to apply and in some respects verges on the metaphysical.  

Why is it significant?

Secondary data, despite its name, is perhaps the most important category of data within the IP Act. It is, roughly speaking, metadata acquired under a targeted, thematic or bulk interception warrant. As such it is not subject to all the usage restrictions that apply to intercepted content.

In particular, unlike for content, there is no requirement to obtain a targeted examination warrant in order to select metadata for examination by use of a selector (such as an e-mail address) referable to someone known to be in the British Islands.

The broader the scope of secondary data, therefore, the more data can be accessed without a targeted examination warrant and the more of what would normally be regarded as content will be included.

Possible interpretations?

Under S.137 of the IPAct secondary data includes:

“identifying data which -

(a) is comprised in, included as part of, attached to or logically associated with the communication (whether by the sender or otherwise),
(b) is capable of being logically separated from the remainder of the communication, and
(c) if it were so separated, would not reveal anything of what might reasonably be considered to be the meaning (if any) of the communication, disregarding any meaning arising from the fact of the communication or from any data relating to the transmission of the communication.”

Identifying data is data which may be used to identify, or assist in identifying, any person, apparatus, system or service, any event, or the location of any person, event or thing.

Identifying data is itself broadly defined. It includes offline as well as online events, such as date or location data on a photograph. However the real challenge is in understanding (c). How does one evaluate the ‘meaning’ of the communication for these purposes? If a name, or a location, or an e-mail address, or a time is extracted from the communication does that on its own reveal anything of its meaning? Is each item extracted to be considered on its own, or are the extracted items of data to be considered together?  How is the ‘meaning’ of a machine to machine communication to be evaluated? Is the test what the communication might mean to a computer or to a human being?

A list of the specific types of data that do and do not fall either side of the line can be a useful aid to understanding abstract data-related definitions such as this. Among the Snowden documents was a GCHQ internal reference list distinguishing between content and related communications data under RIPA.


Applied by or on behalf of

What is the issue?

A technical capability notice (TCN) can require a telecommunications operator to install a specified capability to assist with any interception, equipment interference or bulk acquisition warrant, or communications data acquisition notice, that it might receive in the future.

In particular a TCN can require a telecommunications operator to have the capability to remove electronic protection applied by or on behalf of that operator to any communications or data. This includes encryption. But when is encryption applied "by or on behalf of" that operator?

Why is it significant?

During the passage of the Bill through Parliament there was considerable debate about whether a TCN could be used to stop a telecommunications operator providing end to end encryption facilities to its users. The question was never fully resolved. One issue that would arise, if an attempt were made to use TCNs in that way, is whether the E2E encryption was applied by or on behalf of the operator. If not, then there would be no jurisdiction to issue a TCN in relation to that encryption facility.

Possible interpretations?

In principle, encryption could be applied by the operator, by the user, or by both. An operator would no doubt argue that under the E2E model it is providing the user only with the facility to apply encryption and that any encryption is applied by the user, not the operator.  The strength of that argument could vary depending on the precise technical arrangements in a particular case.


Obtaining data by generation

What is the issue?

The IP Act empowers the Secretary of State, with the approval of a Judicial Commissioner, to give a communications data retention notice to a telecommunications operator. A notice can require the operator to retain specified communications data for up to 12 months.

A data retention notice may, in particular, include:

“requirements or restrictions in relation to the obtaining (whether by collection, generation or otherwise), generation or processing of (i) data for retention, or (ii) retained data.”

This provision makes clear that a requirement to retain data can include obtaining or generating data for retention. But what exactly does that mean? In particular, why does ‘obtaining’ data for retention include ‘generation’?

Why is it significant?

Mandatory communications data retention is one of the most controversial aspects of the IP Act. It is under challenge in the courts and, as a result of previous legal challenges, the government is already having to consult on amendments to the Act.

The powers to require data retention are broader in every respect than those in the predecessor legislation, the Data Retention and Investigatory Powers Act 2014. They can be used against private, not just public, telecommunications operators. They cover a far wider range of data. And they can require data be obtained and generated, not just retained.

So the width of these new powers is significant, especially as telecommunications operators are required not to disclose the existence of data retention notices to which they are subject.

Possible interpretations?

What does it mean to ‘obtain’ data by ‘generation’? It apparently means something different from just generating data for retention, since that is spelt out separately. The most far reaching interpretation would be if the notice could require the operator to require a third party to generate and hand over communications data to the operator. Could that be used to compel, say, a wi-fi operator to obtain and retain a user's identity details?

There was no suggestion during the Parliamentary debates that it could be used in that way, but then the curious drafting of this provision received no attention at all.


‘Internet service’ and ‘internet communications service’

What is the issue?

The IPAct uses both ‘internet service’ and ‘internet communications service’ in its provisions that set out the limits on public authority access to internet connection records (ICRs). However it provides no definitions. Nor are these well understood industry or technical terms.

Why is it significant?

ICRs are logs of visited internet destinations such as websites. ICRs are particularly sensitive since they can be a rich source of information about someone’s lifestyle, health, politics, reading habits and so on. The IP Act therefore places more stringent limits, compared with ordinary communications data, on the authorities that may access ICRs and for what purposes.

The Act stipulates several purposes for which, in various different circumstances, a public authority can access ICRs. They include:
  • to identify which person or apparatus is using an internet service where the service and time of use are already known. (S.62(3))
  • to identify which internet communications service is being used, and when and how it is being used, by a person or apparatus whose identity is already known. (S.62(4)(b)(i) and S.62(5)(c)(i))
  • to identify which internet service is being used, and when and how it is being used, by a person or apparatus whose identity is already known. (S.62(4)(b) (iii) and S.62(5)(c) (iii))

The second and third purposes apply identically to internet services and internet communications services. The first purpose applies only to internet services.

The purposes for which the powers can be used may therefore differ, depending on whether we are dealing with an internet service or an internet communications service. But as already noted, the Act does not tell us what either of these terms means.

Possible interpretations?

We can find clues to interpretation in the footnotes to the draft Communications Data Code of Practice. 

Footnote 49 says that an ‘internet service’ is a service provided over the internet. On the face of it this would seem to exclude a service consisting of providing access to the internet. However the example illustrating S.62(3) in paragraph 9.6 of the draft Code suggests differently.

Footnote 49 goes on to say that 'internet service' includes ‘internet communication services, websites and applications.’ It also suggests examples of online travel booking or mapping services.

This explanation presents some problems.

First is the suggestion that internet communication services are a subset of internet services. If that is right then subsections 62(4)(b)(i) and 62(5)(c)(i) of the Act (above, internet communication services) are redundant, since the respective subsections (iii) already cover internet services in identical terms.

If ‘internet communication service’ is redundant, then the uncertainties with its definition may not signify since S.62 can simply be applied to any 'internet service'.

Elsewhere the draft Code suggests that the subsections (iii) relate to ‘other’ internet services (i.e. additional to internet communications services covered by subsections (i)). However that language does not appear in the Act.

Second is the suggestion that websites and applications are different from internet communications services.  On the face of it an internet communication service could mean just e-mail or a messaging service. But if so, what are we to make of ‘applications’ as something different, since many messaging services are app-based?

Last, to add to the confusion, footnote 48 of the Draft Code of Practice says that an internet communication service is a service which provides for the communication between one or more persons over the internet and ‘may include’ email services, instant messaging services, internet telephony services, social networking and web forums.

This goes wider than just e-mail and messaging services. Does it, for instance, include online games with the ability to chat to other players?  In context does ‘person’ refer only to a human being, or does it include machine communications?

Those involved in authorising and approving applications for access to ICRs will have to take a view on what these terms mean and how they fit together within the scheme of the Act. 

Material whose possession is a crime

What is the issue?

Another ground on which access to ICRs may be obtained is to identify where or when a known person is accessing or running a file or program which “wholly or mainly involves making available, or acquiring, material whose possession is a crime”. There are relatively few offences that are committed by mere possession of material. Illicit drugs and indecent images of children are two mentioned in the draft Code of Practice.

Why is it significant?

The width of the definition affects what kinds of criminal activity can be the subject of applications to access ICRs under this head.

Possible interpretations?

Does the section apply more widely than mere possession, for instance where possession is an offence only if it is with a view to some other activity? What about possession offences where possession is not an offence if it is for personal use?


URLs up to the first slash

What is the issue?

It has long been understood that under RIPA the portion of a web address to the right of the first slash is content, but otherwise the URL is communications data. RIPA contained a convoluted definition designed to achieve that result. Although the Home Office says that the IPAct achieves the same result, exactly how the definitions achieve that is not always obvious.

Why is it significant

Communications data retention and acquisition powers can be deployed only against communication data, not content. So it is important to know what is and is not content.  It is especially important for Internet Connection Records, which the Home Office has repeatedly said include top-level web addresses but not page URLs.

In June 2015, in A Question of Trust at paragraph 9.53, David Anderson Q.C. said that the Home Office had provided him with this definition of 'weblogs' (now known as ICRs):

“Weblogs are a record of the interaction that a user of the internet has with other computers connected to the internet. This will include websites visited up to the first ‘/’ of its [url], but not a detailed record of all web pages that a user has accessed. This record will contain times of contacts and the addresses of the other computers or services with which contact occurred.”

He went on:

"Under this definition a web log would reveal that a user has visited e.g. or, but not the specific page."

He also noted  that:

"Under the current accepted distinction between content and CD, would be communications data while would be content; and this is set out in the Acquisition Code. However there are arbitrary elements to that definition – for example (no ‘www.’) takes you to the same place as”

Possible interpretations

The House of Commons Science and Technology Committee criticised the data definitions in the draft Bill.  They remain complex and abstract in the final legislation.

Towards the end of the pre-Bill scrutiny the Home Office submitted evidence to the Joint Committee that gave more information about what kinds of data would constitute communications data and ICRs. 

In the table at Annex A para 20 of its written evidence the Home Office classified as ‘content’ the following:

“The url of a webpage in a browsing session (e.g. or or friend’”

The first example reflected the prior understanding that a full URL is content. The second and third examples (subdomains) depart from the previous understanding set out in the above extract from ‘A Question of Trust’ by classifying the material to the left of the first slash as content.

Whatever the merits of this approach in removing some of the arbitrariness noted by David Anderson, it is difficult to find anything in the legislation that draws the line at the point suggested. The Home Office evidence gave no explanation of why it drew the line where it did. 

The draft Communications Data Code of Practice does not address the point specifically, but its explanation of fully qualified domain names at page 17 might perhaps suggest that the Home Office has now reverted to the original position described in A Question of Trust.

Given the sensitivity of ICRs this is an area in which clarity is important, not just for ISPs who are subject to the IPAct's requirements but also so that the general public can know what kinds of data are potentially subject to retention and access. 

This is another example pointing to the desirability of publishing a comprehensive list of datatypes illustrating what kinds of data fall into which categories and, by reference to the definitions in the IPAct itself, why they do so.