Lord Allan of Hallam:

Moved by Lord Allan of Hallam

228: Clause 173, page 151, leave out lines 1 and 2Member’s explanatory statementThis amendment removes a requirement on providers which could encourage excessive content removal in borderline cases of illegality.

My Lords, we are coming to some critical amendments on a very important issue relatively late in the Bill, having had relatively little discussion on it. It is not often that committees of this House sit around and say, “We need more lawyers”, but this is one of those areas where that was true.

Notwithstanding the blushes of my noble friend on the Front Bench here, interestingly we have not had in our debate significant input from people who understand the law of freedom of expression and wish to contribute to our discussions on how online platforms should deal with questions of the legality of content. These questions are crucial to the Bill, which, if it does nothing else, tells online platforms that they have to be really robust in taking action against content that is deemed to be illegal under a broad swathe of law in the United Kingdom that criminalises certain forms of speech.

We are heavy with providers, and we are saying to them, “If you fail at this, you’re in big trouble”. The pressure to deal with illegal content will be huge, yet illegality itself covers a broad spectrum, from child sexual exploitation and abuse material, where in many cases it is obvious from the material that it is illegal and there is strict liability—there is never any excuse for distributing that material—and pretty much everyone everywhere in the world would agree that it should be criminalised and removed from the internet, through to things that we discussed in Committee, such as public order offences, where, under some interpretations of Section 5 of the Public Order Act, swearing at somebody or looking at them in a funny way in the street could be deemed alarming and harassing. There are people who interpret public order offences in this very broad sense, where there would be a lot less agreement about whether a specific action is or is not illegal and whether the law is correctly calibrated or being used oppressively. So we have this broad spectrum of illegality.

The question we need to consider is where we want providers to draw the line. They will be making judgments on a daily basis. I said previously that I had to make those judgments in my job. I would write to lawyers and they would send back an expensive piece of paper that said, “This is likely to be illegal”, or, “This is likely not to be illegal”. It never said that it was definitely illegal or definitely not illegal, apart from the content I have described, such as child sexual abuse. You would not need to send that, but you would send the bulk of the issues that we are dealing with to a lawyer. If you sent it to a second lawyer, you would get another “likely” or “not likely”, and you would have to come to some kind of consensus view as to the level of risk you wished to take on that particular form of speech or piece of content.

This is really challenging in areas such as hate speech, where exactly the same language has a completely different meaning in different contexts, and may or may not be illegal. Again, to give a concrete example, we would often deal with anti-Semitic content being shared by anti-anti-Semitic groups—people trying to raise awareness of anti-Semitic speech. Our reviewers would quite commonly remove the speech: they would see it and it would look like grossly violating anti-Semitic speech. Only later would they realise that the person was sharing it for awareness. The N-word is a gross term of racial abuse, but if you are an online platform you permit it a lot of the time, because if people use it self-referentially they expect to be able to use it. If you start removing it they would naturally get very upset. People expect to use it if it is in song lyrics and they are sharing music. I could give thousands of examples of speech that may or may not be illegal depending entirely on the context in which it is being used.

We will be asking platforms to make those judgments on our behalf. They will have to take it seriously, because if they let something through that is illegal they will be in serious trouble. If they misjudged it and thought the anti-Semitic hate speech was being circulated by Jewish groups to promote awareness but it turned out it was being circulated by a Nazi group to attack people and that fell foul of UK law, they would be in trouble. These judgments are critical.

We have the test in Clause 173, which says that platforms should decide whether they have “reasonable grounds to infer” that something is illegal. In Committee, we debated changing that to a higher bar, and said that we wanted a stronger evidential basis. That did not find favour with the Government. We hoped they might raise the bar themselves unilaterally, but they have not. However, we come back again in a different way to try to be helpful, because I do not think that the Government want excessive censorship. They have said throughout the Bill’s passage that they are not looking for platforms to be overly censorious. We looked at the wording again and thought about how we could ensure that the bar is not operated in a way that I do not think that the Government intend. We certainly would not want that to happen.

We look at the current wording in Clause 173 and see that the test there has two elements. One is: “Do you have reasonable grounds to infer?” and then a clause in brackets after that says, “If you do have reasonable grounds to infer, you must treat the content as illegal”. In this amendment we seek to remove the second part of that phrasing because it seems problematic. If we say to the platform, “Reasonable grounds to infer, not certainty”—and it is weird to put “inference”, which is by definition mushy, with “must”, which is very certain, into the same clause—we are saying, “If you have this mushy inference, you must treat it as illegal”, which seems quite problematic. Certainly, if I were working at a platform, the way I would interpret that is: “If in doubt, take it out”. That is the only way you can interpret that “must”, and that is really problematic. Again, I know that that is not the Government’s intention, and if it were child sexual exploitation material, of course you “must”. However, if it is the kind of abusive content that you have reasonable grounds to infer may be an offence under the Public Order Act, “must” you always treat that as illegal? As I read the rest of the Bill, if you are treating it as illegal, the sense is that you should remove it.

That is what we are trying to get at. There is a clear understanding from the Government that their intention is “must” when it comes to that hard end of very bad, very clearly bad content. However, we need something else—a different kind of behaviour where we are dealing with content where it is much more marginal. Otherwise, the price we will pay will be in freedom of expression.

People in the United Kingdom publish quite robust, sweary language. I sometimes think that some of the rules we apply penalise the vernacular. People who use sweary, robust language may be doing so entirely legally—the United Kingdom does not generally restrict people from using that kind of language. However, we risk heading towards a scenario where people post such content in future, and they will find that the platform takes it down. They will complain to the platform, saying, “Why the hell did you take my content down?”—in fact, they will probably use stronger words than that to register their complaint. When they do, the platform will say, “We had reasonable grounds to infer that that was in breach of the Public Order Act, for example, because somebody might feel alarmed, harassed or distressed by it. Oh, and look—in this clause, it says we ‘must’ treat it as illegal. Sorry—there is nothing else we can do. We would have loved to have been able to exercise the benefit of the doubt and to allow you to carry on using that kind of language, because we think there is some margin where you have not behaved in an illegal way. But unfortunately, because of the way that Clause 173 has been drafted, our lawyers tell us we cannot afford to take the risk”.

In the amendment we are trying to—I think—help the Government to get out of a situation which, as I say, I do not think they want. However, I fear that the totality of the wording of Clause 173, this low bar for the test and the “must treat as” language, will lead to that outcome where platforms will take the attitude: “Safety first; if in doubt, take it out”, and I do not think that that is the regime we want. I beg to move.

My Lords, I regret I was unable to be present in Committee to deliver my speech about the chilling effect that the present definition of illegality in the Bill will have on free speech on the internet.

I am still concerned about Clause 173, which directs platforms how to come to the judgment on what is illegal. My concern is that the criterion for illegality, “reasonable grounds to infer” that elements of the content are illegal, will encourage the tech companies to take down content which is not necessarily illegal but which they infer could be. Indeed, the noble Lord, Lord Allan, gave us a whole list of examples of where that might happen. Unfortunately, in Committee there was little support for a higher bar when asking the platforms to judge what illegal content is. However, I have added my name to Amendment 228, put forward by the noble Lord, Lord Allan, because, as he has just said, it is a much less radical way of enhancing free speech when platforms are not certain whether to take down content which they infer is illegal.

The deletion of part of Clause 173(5) is a moderate proposal. It still leaves intact the definition for the platforms of how they are to make the judgment on the illegality of content, but it takes out the compulsory element in this judgment. I believe that it will have the biggest impact on the moderation system. Some of those systems are run by machines, but many of the moderation processes, such as Meta’s Facebook, involve thousands of human beings. The deletion of the second part of Clause 173(5), which demands that they take down content that they infer is illegal, will give them more leeway to err on the side of freedom of speech. I hope that this extra leeway to encourage free speech will also be included in the way that algorithms moderate our content.

Earlier in the Bill, Clause 18 lays out, for all services, the importance of protecting users’ rights to freedom of expression, and there are various duties of assessment for large companies. However, there is not enough in the Bill which builds freedom of expression into the moderation capacity of the platforms. Alan Rusbridger, a member of the Facebook Oversight Board, gave evidence to the communications Select Committee inquiry into freedom of expression online. He said:

“I believe that freedom of speech is a hugely important right … In most judgments, I begin by thinking, ‘Why would we restrict freedom of speech in this particular case?’”.

Evidence was also given that many moderators do not have a background in freedom of expression and are not completely conversant with the Article 10 rights. The amendment will allow moderators to think more about their role in erring on the side of freedom of expression when deciding on the illegality of content.

There has been much discussion, both in Committee and on Report, on protecting freedom of expression, but not much movement by the Government. I hope that the Minister will use this small amendment to push for draft codes of practice which allow the platforms, when they are not sure of the illegality of content, to use their discretion and consider freedom of expression.

My Lords, it is all quite exciting now, is it not? I can say “hear, hear!” a lot; everyone is talking about freedom of expression. I cannot tell noble Lords how relieved and pleased I was both to hear the speeches and to see Amendment 228 from the noble Lord, Lord Allan of Hallam, and the noble Viscount, Lord Colville of Culross, who both explained well why this is so important. I am so glad that, even late in our discussions on Report, it has returned as an important issue.

We have already discussed how in many cases, especially when it comes to what is seen as illegal speech, decisions about illegality are very complicated. They are complicated in the law courts and offline, and that is when they have the full power of lawyers, the criminal justice system and so on trying to make decisions. Leaving it up to people who, through no fault of their own, are not qualified but who work in a social media company to try to make that decision in a climate of quite onerous obligations—and having phrases such as “reasonable grounds to infer”—will lead to lawful expression being overmoderated. Ultimately, online platforms will use an abundance of caution, which will lead to a lot of important speech—perfectly lawful if not worthy speech; the public’s speech and the ability to speak freely—being removed. That is not a trivial side issue; it will discredit the Bill, if it has not done so already.

Whenever noble Lords make contributions about why a wide range of amendments and changes are needed—particularly in relation to protecting children, harm and so on—they constantly tell us that the Bill should send an uncompromising message. The difficulty I have is with the danger that the Bill will send an uncompromising message that freedom of expression is not important. I urge the Minister to look carefully at the amendment, because the message should be that, while the Bill is trying to tackle online harm and to protect children in particular—which I have never argued against—huge swathes of it might inadvertently silence people and deprive them of the right to information that they should be able to have.

My Amendment 229—I am not sure why it is in this group, but that is nothing new in the way that the groupings have worked—is about lawful speech and about what content is filtered by users. I have already argued for the replacement of the old legal but harmful duty, but the new duty of user empowerment is welcome, and at face value it puts users in the driving seat and allows adults to judge for themselves what they want and do not want to see. But—and it is a large but—that will work only if users and providers agree about when content should be filtered and what content is filtered.

As with all decisions on speech, as I have just mentioned, in the context particularly of a heightened climate of confusion and sensitivity regarding identity politics and the cancel-culture issues that we are all familiar with, there are some problems with the way that things stand in the Bill. I hope I am using the term “reasonable grounds to infer” in a better way than it is used in terms of illegality. My amendment specifies that companies need to have reasonable grounds to infer that content is abusive or inciting hatred when filtering out content in those user empowerment tools. Where a user chooses to filter out hateful content based on race, on being a woman or whatever, it should catch only content that genuinely falls under those headings. There is a risk that, without this amendment, technologies or individuals working for companies could operate in a heavy-handed way in filtering out legitimate content.

I shall give a couple of examples. Say that someone chooses to filter out abusive content targeting the protected characteristic of race. I imagine that they would have a reasonable expectation that that filter would target aggressive, unpleasant content demeaning to a person because of their race, but does the provider agree with that? Will it interpret my filtering choice as a user in the most restrictive way possible in a bid to protect my safety or by seeing my sensibilities as having a low threshold for what it might consider to be abuse?

The race issue illustrates where we get into difficulties. Will the filterers take their cue from the document that has just been revealed, which was compiled by the Diocese of St Edmundsbury and Ipswich, which the anti-racist campaigning group Don’t Divide Us has just released, and which is being used in 87 schools? Under the heading of racism we have ideas like passive racism includes agreeing that

“There are two sides to every story”,

or if you deny white privilege or if you start a sentence saying, “Not all white people”. “Veiled racism” in this document—which, as I say, is being used in schools for that particular reason by the Church of England—includes a “Euro-centric curriculum” or “cultural appropriation”. “Racist discrimination” includes “anti- immigration policies”, which, as I pointed out before, would indicate that some people would call the Government’s own Bill tonight racist.

The reason why I mention that is that you might think, “I am going to have racism filtered out”, but if there is too much caution then you will have filtered out very legitimate discussions on immigration and cultural appropriation. You will be protected, but if, for example, the filterer follows certain universities that have deemed the novels of Walter Scott, the plays of William Shakespeare or Enid Blyton’s writing as racist, then you can see that we have some real problems. When universities have said there is misogynistic bullying and sexual abuse in “The Great Gatsby” and Ovid’s “Metamorphoses”, I just want to make sure that we do not end up in a situation where there is oversensitivity by the filterers. Perhaps the filtering will take place by algorithm, machine learning and artificial intelligence, but the EHRC has noted that algorithms just cannot cope with the context, cultural difference and complexity of language within the billions of items of content produced every day.

Amendment 229 ensures that there is a common standard—a standard of objective reasonableness. It is not perfect at all; I understand that reasonableness itself is open to interpretation. However, it is an attempt to ensure that the Government’s concept of user empowerment is feasible by at least aspiring to a basic shared understanding between users and providers as to what will be filtered and what will not, and a check against providers’ filter mechanisms removing controversial or unpopular content in the name of protecting users. Just as I indicated in terms of sending a message, if the Government could indicate to the companies that rather than taking a risk-averse attitude, they had to bear in mind freedom of expression, not be oversensitive and not be too risk-averse or overcautious, we might begin to get some balance. Otherwise, an awful lot of lawful material will be removed that is not even harmful.

My Lords, I support Amendment 228. I spoke on this issue to the longer amendment in Committee. To decide whether something is illegal without the entire apparatus of the justice system, in which a great deal of care is taken to decide whether something is illegal, at high volume and high speed, is very worrying. It strikes me as amusing because someone commented earlier that they like a “must” instead of a “maybe”. In this case, I caution that a provider should treat the content as content of the kind in question accordingly, that something a little softer is needed, not a cliff edge that ends up in horrors around illegality where someone who has acted in self-defence is accused of a crime of violence, as happens to many women, and so on and so forth. I do not want to labour the point. I just urge a gentle landing rather than, as it is written, a cliff edge.

My Lords, this has been a very interesting debate. Beyond peradventure my noble friend Lord Allan and the noble Viscount, Lord Colville, and the noble Baroness, Lady Fox, have demonstrated powerfully the perils of this clause. “Lawyers’ caution” is one of my noble friend’s messages to take away, as is the complexities in making these judgments. It was interesting when he mentioned the sharing for awareness’s sake of certain forms of content and the judgments that must be taken by platforms. His phrase “If in doubt, take it out” is pretty chilling in free speech terms—I think that will come back to haunt us. As the noble Baroness, Lady Fox, said, the wrong message is being delivered by this clause. It is important to have some element of discretion here and not, as the noble Baroness, Lady Kidron, said, a cliff edge. We need a gentler landing. I very much hope that the Minister will land more gently.

My Lords, this has been a good debate. It is very hard to see where one would want to take it. If it proves anything, it is that the decision to drop the legal but harmful provisions in the Bill was probably taken for the wrong reasons but was the right decision, since this is where we end up—in an impossible moral quandary which no amount of writing, legalistic or otherwise, will get us out of. This should be a systems Bill, not a content Bill.

My Lords, I start by saying that accurate systems and processes for content moderation are crucial to the workability of this Bill and keeping users safe from harm. Amendment 228 from the noble Lord, Lord Allan of Hallam, seeks to remove the requirement for platforms to treat content as illegal or fraudulent content if reasonable grounds for that inference exist. The noble Lord set out his concerns about platforms over-removing content when assessing illegality.

Under Clause 173(5), platforms will need to have reasonable grounds to determine whether content is illegal or a fraudulent advertisement. Only when a provider has reasonable grounds to infer that said content is illegal or a fraudulent advertisement must it then comply with the relevant requirements set out in the Bill. This would mean removing the content or preventing people from encountering it through risk-based and proportionate systems and processes.

Clause 173(6) further clarifies what “reasonable grounds to infer” means in relation to judgments about illegal content and fraudulent adverts. It sets out the tests that a provider must apply to the assessment of whether all the elements of an offence—including the mental elements—are present, and whether a defence might be relied on.

The noble Lord’s amendment removes this standard for judging the illegality of content but does not replace it with another standard. That would mean that the Bill provided less detail about when providers are required to treat content as illegal or a fraudulent advert. The result would be that the Bill did not set out a consistent approach to identifying and removing such content that would enable providers to interpret their duties in a broad range of ways while still complying with the framework. This could result in services both over-removing and under-removing content.

I know that the noble Lord is concerned that this provision could encourage overzealous removal of content, but the Government are clear that the approach that I have just outlined provides the necessary safeguards against platforms over-removing content when complying with their duties under the Bill. The noble Lord asked for a different standard to be associated with different types of criminal offence. That is, in effect, what we have done through the distinction that we have made between priority and non-priority offences.

To assist services further, Ofcom will be required to provide guidance on how it judges the illegality of content. In addition, the Government consider that it would not be right to weaken the test for illegal content by diluting the content moderation provisions in the way that this amendment would. Content moderation is critical to protecting users from illegal content and fraudulent advertisements.

The noble Viscount, Lord Colville, set out the importance of freedom of expression, as other noble Lords—principally the noble Baroness, Lady Fox, and the noble Lord, Lord Moylan, but others too—have throughout our scrutiny of the Bill. Our approach regarding freedom of expression recognises that the European Convention on Human Rights imposes obligations in relation to this on states, not private entities. As we have discussed previously, private actors, including service providers in scope, have their own freedom of expression rights. This means that platforms are free to decide what content should be allowed on their sites within the bounds of the law. As such, it is more appropriate to ask them to have particular regard to these concepts rather than to be compliant or consistent with them.

In-scope companies will have to consider and implement safeguards for freedom of expression when fulfilling their duties. For example, platforms could safeguard freedom of expression by ensuring that human moderators are adequately trained to assess contextual and linguistic nuance—such as the examples that the noble Lord gave—to prevent the over-removal of content. The larger services will also have additional duties to assess their impact on freedom of expression and privacy when adopting safety policies, to keep this assessment up to date and to demonstrate that they have taken positive steps in relation to the impact assessment.

Further, platforms will not be penalised for making the wrong calls on pieces of illegal content. Ofcom will instead make its judgments on the systems and processes that platforms have in place when making these decisions. The focus on transparency through the Bill’s framework and on user reporting and redress mechanisms will enable users to appeal the removal of content more effectively than they can at present.

Amendment 229 in the name of the noble Baroness, Lady Fox, would require providers of category 1 services to apply the user empowerment features required under Clause 12 only to content that they have “reasonable grounds to infer” is user empowerment content. The Bill’s cross-cutting freedom of expression duties already prevent providers overapplying user empowerment features or adopting an inconsistent or capricious approach; Ofcom can take enforcement action if they do this. Clause 173(2) and (3) already specify how providers must make judgments about the status of content, including judgments about whether content is in scope of the user empowerment duties. That includes making this judgment based on

“all relevant information that is reasonably available to a provider”.

It is unclear whether the intention of the noble Baroness’s amendment is to go further. If so, it would be inappropriate to apply the “reasonable grounds to infer” test in Clause 173(5) and (6) to user empowerment content. This is because, as I have just outlined in relation to the amendment in the name of the noble Lord, Lord Allan, the test sets out the approach that providers must take when assessing whether content amounts to a criminal offence. The test cannot sensibly be applied to content covered by the user empowerment duties because such content is not illegal. It is not workable to suggest that providers need to apply criminal law concepts such as intent or defences to non-criminal material. Under Clause 48, Ofcom will be required to produce and publish guidance that sets out examples of the kinds of content that Ofcom considers to be relevant to the user empowerment duties. This will assist providers in determining what content is of relevance to the user empowerment duties.

I hope that this allays the concerns raised by the noble Baroness and the noble Lord, and that the noble Lord will be content to withdraw his amendment.

My Lords, I remain concerned that people who use more choice words of Anglo-Saxon origin will find their speech more restricted than those who use more Latinate words, such as “inference” and “reasonable”, but the Minister has given some important clarifications.

The first is that no single decision could result in a problem for a platform, so it will know that it is about a pattern of bad decision-making rather than a single decision; that will be helpful in terms of taking a bit of the pressure off. The Minister also gave an important clarification around—I hate this language, but we have to say it—priority versus primary priority. If everything is a priority, nothing is a priority but, in this Bill, some things are more of a priority than others. The public order offences are priority offences; therefore, they have a little bit more leeway over those offences than they do over primary priority offences, which include the really bad stuff that we all agree we want to get rid of.

As I say, I do not think that we are going to get much further in our debates today although those were important clarifications. The Minister is trying to give us reasonable grounds to infer that the guidance from Ofcom will result in a gentle landing rather than a cliff edge, which the noble Baroness, Lady Kidron, rightly suggested is what we want. With that, I beg leave to withdraw the amendment.

Amendment 228 withdrawn.

Amendment 229 not moved.