Supplementary written evidence submitted by Logically (SMH0076)

Would full implementation of the Online Safety Act 2023 have helped prevent the spread of misinformation that fuelled the violent disorder that took place across the United Kingdom following the tragic murders in Southport on 29 July 2024?

1. The Online Safety Act 2023 (“OSA”) imposes a number of duties on providers of regulated user-to-user services and search services (collectively referred to hereinafter as “platforms”). These include duties to take proportionate measures to prevent and mitigate users’ exposure to “illegal content” on their services.[1]
2. The OSA imposes specific duties on regulated platforms to proactively protect their users from content relating to “priority offences”. These are the most serious kinds of harmful online content, including those relating to terrorism, child sexual exploitation and abuse, and other offences specified in Schedule 7 of the OSA. One of the priority offences identified in Schedule 7 of the OSA is “foreign interference”, as defined in Section 13 of the National Security Act (“NSA”).
3. User-to-user services like Meta’s Facebook or X (formerly Twitter)l need to ensure that they take proportionate measures to prevent users from encountering content related to priority offences; mitigate and manage the risk of their platform being used to facilitate priority offences; and minimise how long content related to priority offences remain on their platform.[2] Platforms also have to conduct risk assessments and factor these into the mitigation measures they apply.[3]
4. Prior to the coming into force of the OSA, many of the larger user-to-user services had already put in place processes which could, arguably, help them fulfill their obligations under the new legislation to prevent users from encountering content relating to priority offences. The picture is less clear for smaller user-to-user services or certain kinds of decentralised large user-to-user services like Reddit, as well as search services. Large search services like Google and Bing did have certain measures in place which could help prevent some content relating to priority offences being shared with their users (for instance, through policies for advertisements) but this was not consistent for all kinds of priority offences.
5. The unrest that followed the tragic murders in Southport demonstrated that even when platforms had processes in place, these were not always effective. In our previously submitted evidence to the House Affairs select committee on the Summer 2024 disorder,[4] we drew attention to examples of this across multiple platforms, demonstrating that this was not a problem limited to any specific platform:
1. The false name of the attacker “Ali al-Shakati” – even after this was debunked – featured as a “Trending in the UK” topic suggested to X users under the “What’s happening” sidebar, and users who searched for “Southport” on X were recommended results from users calling to “remove Islam from Britain. Completely and entirely”.[5]
2. Recommender systems such as Meta’s continued to promote these posts even after the police had confirmed the name of the perpetrator was false, and despite users flagging that many posts were in breach of platforms’ terms and conditions with regard to hate speech and inciting violence.[6] Meta’s Oversight Board itself recognised this and has been looking into it.
3. On TikTok, specific accounts were set up that posted locations of violent protests, calling for support and using far-right symbols and messages of ‘mass deportation’ — screenshots from these accounts were also cross-posted to other social media platforms and also to Telegram.[7]
6. It is our belief that a fully implemented OSA and attendant framework (including guidance from Ofcom) can help refine and improve pre-existing processes adopted by platforms, while also requiring platforms which did not have processes in place to take action. The requirement under the law for proactive measures could also ensure that the platforms are running proactive, always-on systems to tackle illegal content affecting public health, public safety and national security, rather than rigorously enforcing these systems only after critical incidents or around reporting times, as has sometimes been the case in practice.
7. The OSA’s requirements for risk assessments would also arguably help ensure that processes put in place by platforms to protect users from priority offences and mitigate risks are, to the best extent possible, fit for purpose. The regular cadence of risk assessments can also ensure that the platforms are protected against evolving threats, including the use of generative artificial intelligence and other emerging technologies by bad actors.
8. As noted in our previous submission, it would also be useful for the government to retrospectively investigate incidents of misinformation and disinformation like the Southport unrest. Not only will this provide a better understanding of how the OSA can help – and where it may need to be improved upon – it should also increase public awareness of illegal online behaviour and how to protect themselves from falling prey to it.
9. We reiterate that there is a clear case for Ofcom to ensure their guidance sets out clear thresholds for when platforms must take action to tackle content which amounts to priority offences like foreign interference. For instance, when it comes to the foreign interference offence, Ofcom’s guidance suggests that there are no generic examples of what this would look like in practice, and that there is no robust data on the reach of such efforts among UK internet users.
10. In addition to the existing guidance’s reference to the use of bots, however, it is possible for Ofcom and the Government to prepare generic behavioural profiles that can be associated with foreign interference, building on tactics, techniques and procedures (TTPs) identified in credible expert frameworks, including the DISARM Framework[8] or the “kill chain” framework.[9] The European External Action Service (EEAS) uses the DISARM Framework in its analytical approach to foreign information manipulation and interference, as they consider it to be state of the art,[10] and its TTPs have even been adopted in the EU’s strengthened Code of Practice on Disinformation. 
11. Similarly, the guidance could also include more detail on how to assess the impact of influence operations. Experts and researchers have developed solutions such as the Brookings Institute’s “Breakout Scale”, which measures the impact of covert influence operations on a scale of 1 (lowest) to 6 (highest). The Breakout Scale was recently used by Open AI[11] to measure the impact of several influence operations it discovered that were seeking to use its models and tools. The influence operations were scored 2 on the Breakout Scale, which indicated that they were able to spread to multiple platforms, but were unable to ‘breakout’ to authentic audiences.
12. We also believe that Ofcom should consider updating their guidance with some of the lessons learned from the Southport unrest. The updates could include case studies on how content relating to various priority offences was disseminated and amplified online, abetting and facilitating the violent disorder on the streets. A particularly useful case study could be on foreign interference and the role of information laundering across overt and covert state-backed campaigns. In the case of Southport, this could include examining the amplification of content from  fringe ‘news’ website “Channel 3NOW” by Russian media and Russia-linked actors.[12] Arguably, elements of this content met the threshold for platforms to take proactive measures to curb circulation under the OSA in that there were ‘reasonable grounds to infer’ that it was illegal.
13. While platforms need to learn lessons from the Southport unrest regarding where their processes failed, the regulator should also consider how its guidance can be made more effective in light of it. Such exercises should follow after subsequent events or incidents of a similar nature as well, and could be conducted with the involvement and advice of the Advisory Committee on Disinformation and Misinformation set up under Section 152 of the OSA.

Should platforms employ third party fact checking services by expert organisations or community-led moderation efforts to protect users from misinformation?

14. The International Fact-Checking Network (IFCN) was founded nearly ten years ago and helped bring greater professionalism, trust and accountability to the global fact checking community with its standardised Code of Principles. This in turn allowed the community to play a significant role in public discourse, with platforms such as Meta creating third party fact checking (“3PFC”) progammes that utilised the expertise of professional fact checking organisations to advise content moderation.
15. While this model helped address misinformation and disinformation relating to elections, public health, military conflicts, natural disasters and other high-risk matters over the last decade, the contemporary information environment has evolved in significant ways during this time. It is faster and more algorithmic — fueled by AI-generated content, sophisticated disinformation networks, and virality-driven platforms. Contemporary influence operations attempt to do more than just spread false or misleading information about a single issue or event; they seek to achieve broader strategic objectives, which include the undermining of democratic societies by targeting public institutions, democratic processes and expert knowledge.
16. Fact checking, as it was initially designed, cannot always keep pace with the scale of today’s online manipulation tactics. The IFCN model is built on the idea of credibility around a code of principles, ensuring that fact-checking organizations followed rigorous methodologies. But this traditional model relies too heavily on slow, manual processes — it was built for a time when misinformation spread article by article, not at AI speed. There are also questions to be asked of the financial sustainability of this model, which for many organisations has meant reliance on the largesse of platforms.
17. While there is a need to update this model, including scaling it with the use of technology, this does not mean it should be abandoned in favour of user-generated, crowdsourced solutions like community notes – currently employed by X and soon to be employed in the US by Meta. 
18. While community notes-style models can be less expensive and offer ostensibly high scalability, there are reasons to be cautious about their adoption. These solutions rely on crowdsourced consensus, meaning that fact-based corrections may fail to gain traction if they contradict dominant narratives or become the target of brigading efforts. This makes the system more vulnerable to influence and coordinated behaviour, introducing the need for stronger safeguards against coordinated misinformation campaigns.
19. When it comes to high-stakes issues like public health, election integrity and security, it is also risky to leave fact checking to what may devolve into popularity contests or crowdsourced opinions. Community notes-style models can introduce diverse perspectives, but having various viewpoints is not the same as possessing the expertise needed to assess the accuracy of information.
20. Instead of relying solely on either a professional third party fact checking model or a community notes-style model, the modern information environment would be better served by  a hybrid ecosystem where AI, platforms, fact-checkers, and users enrolled in community notes models work together to verify content at speed and scale, focusing on harm and virality.
21. As noted in our previous evidence to this inquiry,[13] there are a growing number of technological solutions providers focused on information integrity in the market, and the UK is increasingly finding a foothold in this space. The AI-enabled systems developed by these providers can be used to identify and triage misinformation based on its potential harm and virality, allowing accredited organizations and experts to intervene before falsehoods gain traction. Emerging narratives can be segmented based on risk levels, ensuring that high-impact claims receive expert attention, while lower-risk misinformation is managed through scalable community-driven mechanisms. For instance, content relating to terrorist attacks or other serious crimes or serious public health emergencies should be considered high risk, and require assessment by experts. On the other hand, debates over simple historical facts (like when a particular law was passed or a ruler was crowned or a battle was fought) can be left to be handled by community notes.
22. Figure 1 below describes how this hybrid, risk based approach could operate in practice.

Figure 1: Proposed model for hybrid risk-based approach to content moderation

23. In addition to the risk-based approach, the fact checks or other resulting content meant to counter misinformation and disinformation must move beyond static articles and instead be embedded directly within content streams, search results, and conversational AI. This ensures that corrections are contextual, frictionless, and surfaced where misinformation spreads. Recent efforts by BlueSky could offer a useful template, including options for users to subscribe to moderation of key harms they are concerned about – which also helps ensure that users feel more trust in the system. Such dynamic content could also utilise technology-driven solutions like counterspeech,[14] which research has proven to be an effective method for mitigating online hate while maintaining a diversity of voices and opinions.[15]
24. The adoption of such hybrid models, including the involvement of professional fact checking organisations for high-risk mis-and dis-information, can be encouraged through guidance issued by Ofcom on implementation of the OSA, or even subsequent versions of its codes of practice. 

3 April 2025