Could public debate on immigration be suppressed by the Online Safety Bill?

A little bombshell, hidden deep within the gargantuan piece of legislation that is the Online Safety Bill, could have the effect of suppressing public debate around immigration. This the phrase is “Assisting illegal immigration” and it refers to the criminalisation of people helping refugees to arrive in the UK. It links the Online Safety Bill to the Nationality and Borders Act in such a way that it creates a censorship requirement on social media platforms. Public discourse around safe and legal routes is most likely to be impacted.

The Online Safety Bill imports Section 41 from the Nationality and Borders Act. It does so by means of a reference to the 1971 Immigration Act – Section 25 Assisting Unlawful Immigration – which is amended by Section 41 of the Nationality and Borders Act. The reference is hidden deep in a Schedule appended to the Bill (Schedule 7). Section 41 is a highly controversial piece of law, part of the so-called “New Plan” introduced by Home Secretary Priti Patel, and furthers the government’s “hostile immigration environment” policy. It establishes a criminal offence of facilitating the arrival of asylum seekers to the UK, with a maximum penalty of life imprisonment.

Social media providers who identify this offence in content posted on their platforms would be under an obligation to remove it or prevent people from seeing it. Potentially, an offending social media post could be made to disappear whilst it is being uploaded by the user – a form of interception and characteristic of prior censorship.

There is a fundamental question about how platform providers go about detecting and identifying this content in a social media post. They could block people smugglers’ accounts, but more likely in the current political climate, they will look for posts containing images of what the government does not want to see – people arriving at UK beaches in small boats – and remove them.

The chances are they would rely on artificial intelligence techniques, or content moderation systems based on perceptual hashing. Both options entail risks of over-blocking. Lawful posts could be censored, with serious implications for public discourse in the UK.

“Content” means “words, images, speech or sounds” either individually or “taken together with other regulated user generated content present on the service” that could constitute an offence. The problem for online platforms is how to evaluate the content to determine whether or not it does comprise an offence.

From the viewpoint of a social media platform, it is far from clear how to interpret content that would be illegal under Section 41. What would a post look like? Platform policing is conducted at scale using automated content moderation systems. They could take the approach of building what’s known as a “hash” database, containing digital fingerprints of previously identified content of, for example, boats full of people crossing the Channel. They would remove any images that match something in the database. However, content related to organised immigration crime can be more complex, and may be context-dependent. To address this, they would likely use AI systems using text-based classifiers. Both system types can have high error rates or false positives.

Providers are asked to evaluate the content and assess where they “reasonably consider” it is illegal. A government amendment (NC14) at Report stage says providers can judge the content “on the basis of all relevant information that is reasonably available”. They are asked to consider whether the mental element of the crime is present. However, one may reasonably question why they are asked to “judge” the content at all.

The Bill is vague on how it expects online platform providers to tackle this. They are given instructions to “prevent individuals from encountering” the illegal content, or to “minimise the time it is present”. This is legal language that directs platforms towards the use of automated content moderation systems and potentially what’s known as an “upload filter”. As with so much in this Bill, it never actually says it specifically, but it’s likely to be what will be required in order to comply.

An “upload filter” intercepts content while the user is uploading it onto the platform, and checks to see if there is a match for illegal content. It operates a form of surveillance on users posting to the platform, 24/7. When it finds a match, it removes the content, effectively making it disappear before the user even knows what has happened. It is prior censorship, also known as prior restraint – and prohibited in English law.

The requirement is extended to search engines. Websites would be demoted in search listings if they have content deemed illegal under the Nationality and Borders Act. This could leave websites in the invidious position of losing their traffic, without a clue as to why it is happening.

None of this is good for free speech. If online platforms are swept clean of these images and posts, a vital area of public debate vanishes from the conversation.

With immigration issues regularly appearing at the top of the news agenda in the UK, it’s a deeply worrying an example of how the Online Safety Bill can be used for the government’s own purposes, to create a surveillance-based form of censorship and suppress debate on vital issues of public policy.

Dr Monica Horten is ORG’s Policy Manager – Freedom of Expression. She is policy lead on the Online Safety Bill.

Stop state censorship of online speech

Support the campaign