Neural Network – October 2026
In this edition of the Neural Network, we look at key AI developments from September and October 2026 and reflect on our third annual AI Afternoon event in London.
In regulatory and government updates, we review the UK Government’s AI Risk Management Toolkit, examine the proposed bill to ban artificial superintelligence in the US and consider proposed restrictions on digital twins in the UK. We also highlight the European Commission’s initiatives to strengthen copyright protection for creative works and innovations and outline some of the notable features of the Supreme People’s Court of China’s new guidelines on handling AI-related legal cases.
In technology developments and market news, the Italian Prime Minister has filed to trademark her own voice.
More details on each of these developments are set out below.
Regulatory and Government Updates
US Senator introduces a bill to ban artificial superintelligence
Parliament considers regulation of AI-generated digital replicas
China’s Supreme People’s Court releases guidance for AI-related legal cases
Technology Development and Market news
AI Afternoon
Our Technology team recently hosted our annual AI Afternoon event in London, exploring the latest developments from regulatory, deployment, and risk perspectives. Joined by external panellists Dee Masters of Cloisters Chambers and Maxime Le Berre of Botifull, the team provided a regulatory update as well as covering topics including AI and its impact on consumers and competition; managing the risks of smart glasses; agentic AI; and the use of AI in litigation.
Some of the key takeaways from the afternoon include:
The obligations under Article 50 of the EU AI Act are now in force, and providers and deployers of AI systems should be complying with their respective transparency obligations.
Agentic AI creates new risks and can make the AI “black box” even deeper. Organisations should fully map any agentic workflows and analyse the legal, regulatory, and contracting risks of the end-to-end agentic AI ecosystem they are operating in.
AI wearables are on the rise, and organisations should consider the risks associated with their use and update their policies accordingly. Generally speaking, consumer AI products are not compatible with enterprise use.
AI is being deployed in litigation with mixed results, fuelling claims by litigants in person and creating new disclosure requirements. Organisations should ensure their people are appropriately trained to use AI responsibly and remain mindful that AI prompts could be disclosable and may waive privilege.
As the regulatory landscape continues to evolve, we have launched a new AI Resources platform, designed to support organisations navigating regulatory and compliance challenges, manage risk and unlock the potential of AI. This platform features a curated selection of our content and practical tools on AI.
Regulatory and Government updates
UK Government publishes AI Risk Management Toolkit
On 8 September 2026, the Department for Science, Innovation and Technology published an AI Risk Management Toolkit (the “Toolkit”) and accompanying workbook. The Toolkit is intended to guide and support the design, operation and delivery process of AI-enabled products, particularly in the public sector. It encourages public bodies to keep a log of AI risks and share them with the UK Government’s Central AI Risk Toolkit team.
The Toolkit is designed to work alongside the Cyber Assessment Framework, a high-level framework produced by the National Cyber Security Centre to ensure consistent assessment of the cyber resilience of organisations operating essential services.
Although the Toolkit is mainly designed to support AI adoption within UK government and the wider public sector, the guidance can be useful when assessing and managing risks across AI products, particularly in terms of compliance with relevant regulations. Specifically, Appendix 1 details a set of risk identification questions which can help organisations flag inconsistencies in their processes with relevant laws and regulations.
The Toolkit includes guidance and risk identification measures and frameworks, alongside a monitoring dashboard, and covers nine risk categories:
Legal, regulatory and compliance – organisations are directed to consider the risks of failing to comply with legal and regulatory requirements relevant to the specific AI product, for example data protection, equality and AI-specific laws.
Accountability and governance - organisations should assess the risks associated with liability for AI outcomes, as well as potential failures in oversight and inadequate monitoring processes.
Contestability and redress – organisations are advised to consider users’ ability to challenge or appeal decisions made by AI systems, the absence of accessible mechanisms for redress, and the difficulties in correcting errors or unfair outcomes resulting from AI decisions.
Appropriate transparency and explainability - organisations are directed to consider how explainable the system’s decision process is and the means by which they comply with obligations to inform users and/or regulators about the AI’s logic and outcomes.
Fairness – organisations are encouraged to review risks to fairness, including biased outcomes, discrimination, the creation of unfair market conditions, and the reinforcement of existing inequalities or exclusion of vulnerable individuals.
Financial – organisations are prompted to consider the likelihood of additional and unexpected costs, losses arising from system errors or failures, and the impact of regulatory fines or compensation claims arising from the AI system’s operation.
Security – organisations are directed to consider their vulnerabilities to cyberattacks, data breaches, unauthorised access to sensitive data, exploitation by malicious actors, and threats to the integrity and reliability of the AI system.
Technical robustness - organisations should assess the risk of system failures and inaccurate outputs, including whether the system has undergone sufficient testing, potential lack of resilience to unexpected inputs, and possible challenges in maintaining or updating the system.
Environment and people – organisations are directed to consider risks relating to negative environmental impacts and negative impacts on people/citizens, such as job displacement, harm to wellbeing, or unintended consequences affecting citizens and communities.
The Toolkit also sets out several types of risk associated with harnessing AI, assigning each an “Impact Level” from negligible to catastrophic based on its impact in relation to the risk appetite of the organisation. It considers business, staff, service/user, reputational impacts, and assists organisations with quantifying (where it is a quantifiable risk) the potential risk posed for different use cases.
US Senator introduces a bill to ban artificial superintelligence
On 23 September 2026, Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act (the “Bill”), formalising a proposal they first announced on 3 September 2026. The Bill proposes to ban the development or deployment of artificial superintelligence, including systems that surpass human intelligence, have the potential to overthrow human governments or have dangerous abilities such as subverting shutdown commands. It also proposes to pause advanced AI development until a new federal regulator is operational and has set safety rules and a model review process.
In our June edition, we covered a US executive order seeking early government access to advanced AI models, which expressly stopped short of mandatory licensing or approval. The Sanders/Casar proposal goes further, as it would place oversight with a new cabinet-level Department of Artificial Intelligence, require developers to submit plans before development, obtain approval before public deployment, and allow the department to supervise the removal of dangerous capabilities. The Bill would also capture systems showing “superintelligence precursor characteristics”, including capabilities to improve the design of nuclear, chemical or biological weapons, deceive humans or modify their own functions. Its enforcement model is deliberately severe, as companies could face dissolution and individuals could face up to 20 years’ imprisonment, while the US would be directed to pursue international agreements and export controls to prevent superintelligence being developed elsewhere.
Whilst the Bill is unlikely to pass in its current form, it is a useful marker of the direction of travel for the wider AI regulatory debate. This sits alongside the more operational EU compliance developments we covered in our August edition and in our article, EU AI Act Update, where we noted the EU’s move from high-level AI governance principles towards concrete obligations for providers and deployers. The Sanders/Casar Bill sits at the more interventionist end of that trend. For organisations that rely on frontier models, the practical point is that there is a chance that access to advanced AI may become more conditional, interruptible and dependent on continuing regulatory approval.
The Bill draws a line between advanced AI, and artificial superintelligence, which would be permanently prohibited. In practice, that boundary is likely to be contested, particularly where a model has some dangerous capabilities but does not obviously exceed human ability across most domains. We will continue to monitor the Bill’s progress.
Parliament considers regulation of AI-generated digital replicas
On 9 September, a new Private Member’s Bill, Personal Data (Digital Twins) Bill (the “Digital Twins Bill”), was introduced in Parliament to require that explicit consent be obtained for the creation of an algorithmic “digital twin”. The Digital Twins Bill’s proponent, Dame Chi Onwurah, describes a digital twin as an automated version of an individual, created by combining personal data with AI. She notes that such digital twins can be used to accurately predict behaviour, target advertising, influence individuals, or impersonate them.
Definition of digital twin
The scope of a “digital twin” in the proposal is broader than the commonly referenced definition, which usually focuses on “images, videos, and audio recordings created by digital technology to realistically replicate an individual’s voice or appearance”; namely deepfakes. The proposed definition would encompass information collected to build up a picture of an individual’s likeness or personality, covering a broader range of software and algorithms that companies are likely already utilising to profile their customers.
Scope of the proposals
The Digital Twins Bill would prohibit processing personal data to build a holistic profile of an individual without their express consent. The UK GDPR already regulates this, requiring a lawful basis, transparency, and appropriate safeguards. However, the Digital Twins Bill is also intended to address malicious deepfakes and to protect individuals in the arts, entertainment, and technology sectors against replicas of their work.
Dame Chi Onwurah clarified the Digital Twins Bill would prohibit the digital twinning of children, recognising that they cannot provide meaningful consent and expressly excluding parental consent as a substitute.
The Digital Twins Bill is scheduled for its second reading in the House of Commons on 13 November 2026. We will continue to monitor its progress, particularly with regard to the publication of the draft wording and will report on any developments.
The European Commission explores targeted copyright reforms
To ensure copyright protection as artificial intelligence (“AI”) continues to advance, the European Commission (the “Commission”) has launched a public consultation, the Targeted Consultation on a Potential Initiative for a Better Copyright Environment for European Creativity and Innovation (the “Consultation”), inviting feedback from copyright holders and other stakeholders across the AI supply chain on the challenges arising from the development of AI as well as other copyright issues.
This initiative follows the European Parliament’s recent resolution, “Copyright and generative artificial intelligence - opportunities and challenges”, which we covered in our March 2026 Edition of Neural Network.
In May 2026, the Commission signalled its intention to consider targeted changes to the EU copyright framework. The current Consultation, which opened on 29 September and runs until 3 November 2026, marks a key step in shaping future policy to manage the complex impact of AI on copyright protection.
The Consultation seeks input on several issues:
The use of copyright-protected material by AI - AI models depend on vast data sets including books, music, images and videos. Advocates have expressed concerns about whether copyright holders are appropriately notified when their works are used and if they are receiving fair compensation.
Combatting online piracy of live events - online piracy remains a significant challenge for live events such as sports, concerts, and broadcasts and the Consultation seeks input on how best to protect the authorised distribution of such material.
Single equitable remuneration right for music performers and producers - amid the rise of streaming platforms and AI-generated music, the Consultation aims to address how to ensure fair compensation for artists and producers when their works are used or adapted by new technologies.
Copyright in Scientific Research - the Consultation examines copyright implications for scientific research, specifically when it is reused, shared and/or used to train an LLM.
The Consultation will be used to inform proposals for legislative reform. We will monitor this closely and provide any updates to copyright legislation as they arise.
China’s Supreme People’s Court releases guidance for AI-related legal cases
On 7 September 2026, the Supreme People’s Court of China published guidance for the adjudication of AI disputes - the Opinions on the Trial of AI-Related Dispute Cases According to Law (the “Opinions”). China has not yet enacted AI-specific legislation; consequently, the Opinions synthesise analysis by the lower courts of existing legislation, including the Civil Code, Civil Procedure, Cybersecurity, Data Security, Copyright, Anti-Unfair Competition, Consumer Rights Protection, and Personal Information Protection Law.
Overall, the Opinions represent a significant development in clarifying China’s approach to AI-related disputes, providing clear guidance in the absence of specific AI legislation. While we have highlighted some of the key points below, the Opinions are much broader in scope. Companies operating in China should consider the Opinions in their entirety to ensure compliance and mitigate the risk of judicial scrutiny.
The Opinions are split into five categories:
(i) Guiding principles
The Opinions require judges to consider three key principles when determining AI-related cases: the well-being and protection of individual rights; the fostering of innovation; and the balance of safety and security. Judges are encouraged to:
be alert to potential discriminatory risks embedded within AI and safeguard the rights of individuals to choose when to be subject to automated decision-making;
ensure fair adjudication to support integration and unlock AI’s potential; and
balance safety and security, with responsibility allocated among developers, providers, and users based on the nature and scale of harms and risks.
(ii) AI tort liability
The Opinions provide guidance on the interaction between numerous torts and AI. A notable development is the approach to establishing whether an AI product is defective. The Opinions explain it is important to consider whether the AI product poses an “unreasonable danger”, assessed by considering the following factors:
the nature and purpose of the AI;
the product’s self-learning ability;
the product’s upgrade and update status;
the user’s control; and
the product’s compliance with relevant national and industry standards
(iii) AI intellectual property disputes
The Opinions direct that liability for copyright infringement should be allocated among developers, providers, and users based on fault. The courts are directed to assess
the party’s contribution;
the source of training data;
infringement-specific preventative measures; and
the profits earned.
(iv) Procedural and evidentiary rules
The Opinions recommend that the courts consider the complexity and opacity associated with evidence gathering in relation to new and emerging technologies. Particularly, the courts are advised to carefully inspect the truthfulness and completeness of digital data generation and collection.
Where AI is used to fabricate facts in civil cases through human intervention - such as alteration, deletion, or manipulation – the courts are empowered to reject claims, impose fines and detention, and pursue criminal liability where appropriate.
(v) Trial mechanisms
Finally, the Opinions advise that the higher courts may assume initial jurisdiction over AI cases that involve significant interests, complexity, have the potential to set a precedent or provide clarity to aid the establishment of uniform legal standards.
Technology Developments/Market News
Italian Prime Minister files to trademark her voice
Italian Prime Minister Giorgia Meloni has applied to the European Union Intellectual Property Office (“EUIPO”) to trademark her voice, to protect herself against AI deepfakes. The application includes a four-second audio clip of Meloni saying “Io sono Giorgia” (“I am Giorgia”) twice. The phrase was the title of her 2021 autobiography and is associated with her political campaigns.
The filing is significant as Meloni appears to be the first major politicians to protect her voice in this way. To date, such measures have primarily been taken by artists and entertainers. The timing is also significant, as the application comes ahead of Italy’s general elections next year. Meloni has previously been targeted by deepfakes, in 2024 she sought €100,000 in damages over deepfake pornographic videos. Earlier this year, she warned: “Deepfakes are a dangerous tool, because they can deceive, manipulate, and strike anyone. I can defend myself. Many others cannot”. Her remarks draw a direct line from her own experience of reputational harm to a systemic, society-wide vulnerability: if a head of government can be targeted, anyone can, and most people lack the resources she has to fight back.
Meloni follows several high-profile figures in taking this step. In January, US actor Matthew McConaughey registered US trademarks covering his image and voice, and in April singer Taylor Swift filed three US trademark applications, including two audio clips of her introducing herself. In the UK, more than 80 performers, including Matt Lucas, Hugh Bonneville and Nicola Coughlan, have called for legislation giving every person a legal right to own their voice.
It is still unclear whether trademark law can protect a person’s voice from AI misuse in practice. Italian media have reported that, even if approved, the trademark would not automatically prevent AI-generated imitations of her voice, although it could provide an additional legal tool against unauthorised use.
The trademark application is still under examination with the EUIPO. The outcome of the application could shape how identity rights develop alongside AI regulation and whether other politicians follow Meloni’s lead. We will continue to monitor the progress and further developments on this application.
