In this edition of our global AI bulletin, we will be looking at:
- Asia
- Hong Kong:
- effects of unsupervised AI-use risk on data privacy and security
- new article published on creating an internal generative AI policy
- Singapore:
- new initiatives and tools launched for AI and data protection
- Europe — Proposal to adhere to the Convention on AI; Commission launches AI tools for researchers and industry; General-Purpose AI Code of Practice published; Guidelines for providers of general-purpose AI models published; European Commission publishes GPAI Training Data Summary Template; European Commission allegedly withdraws AI liability and patents proposals
- UK — Regulators guide firms on responsible AI use; Compliant use of AI in advertising — the UK’s stance; Ofcom’s updates on online safety and AI; AI tools set to transform UK audit practices; How can tech firms help users spot deepfakes?; Compute Roadmap — advancing AI in the UK; Government report on the impact of social media algorithms and generative AI (GenAI)
- US — Department of Commerce renames and reforms AI Safety Institute; Senate decides not to impose AI law moratorium; White House issues AI Action Plan and executive orders; California regulates business use of automated decision-making technology; Colorado to convene special legislative session to consider implementation changes to pioneering AI law; National Institute of Standards and Technology is developing resources to meet demand for additional AI guidance; Lawmakers in Texas enact the Texas Responsible Artificial Intelligence Governance Act
- Global — Harmonized AI standards to reduce fragmented global rules
Hong Kong: Effects of unsupervised AI-use risk on data privacy and security
On June 9, 2025, the Privacy Commissioner for Personal Data (PCPD) published a letter emphasizing the need for AI policies in organizations. The PCPD states that using generative AI without guidance risks personal data privacy and may compromise the organization’s interests. The PCPD has also published a checklist to help organizations create internal AI guidelines.
The checklist covers five key areas: permitted AI use, data privacy, ethical use, data security, and policy violations.
Impact: Organizations should specify which generative AI tools employees may use and for what purposes. These could include drafting documents, summarizing information, or generating advertising materials. The PCPD aims to ensure compliance with the Personal Data (Privacy) Ordinance and protect both organizational and personal data interests.
Hong Kong: New article published on creating an internal generative AI policy
On June 17, 2025, the Office of the Privacy Commissioner for Personal Data (PCPD) issued an article to help organizations develop internal generative AI (GenAI) policies. While GenAI can boost productivity, it may also pose challenges in ensuring data security and protecting personal data privacy for organizations. The article:
- addresses data privacy, lawful use, and data security
- recommends specifying permissible GenAI tools, defining their use, and ensuring compliance with the Personal Data (Privacy) Ordinance
- emphasizes preventing bias and ensuring ethical AI use
Impact: Organizations are encouraged to:
- provide clear instructions on data inputs and outputs and establish policies for data retention and security
- devise an internal AI policy to provide clear guidance to employees on the use of GenAI at work
- train employees, enhance transparency, and establish feedback mechanisms
This approach aims to create a win-win situation by leveraging AI benefits while safeguarding organizational and employee interests.
Singapore: New initiatives and tools launched for AI and data protection
On July 7, 2025, the Infocomm Media Development Authority (IMDA) launched three initiatives to support trusted AI and data use. The initiatives are:
- an expanded Global AI Assurance Sandbox (Sandbox), which helps firms test AI responsibly, including agentic AI and risks like data leakage
- a Privacy Enhancing Technologies (PETs) adoption guide, which offers tools and checklists to help businesses adopt privacy-preserving technologies
- a new national Data Protection Trustmark (DPTM) Singapore Standard aligning with global benchmarks and offering businesses clarity around data protection requirements
Impact: The initiatives aim to balance innovation with strong safeguards and public trust. The Sandbox will also be open to sector regulators who want to develop and get real-life feedback on their AI governance/testing guidelines. Insights will inform policy guidance, with the aim of developing accreditation for testers. Businesses are invited to contribute feedback on the PETs guide (by sending an email to data_innovation@imda.gov.sg) to help refine it as a “living document.” Businesses demonstrating accountable data protection practices can now apply to be certified under the new DPTM Singapore Standard.
EU: Proposal to adhere to the Convention on AI
On June 3, 2025, the European Commission (EC) published a proposal for a decision to initiate the conclusion of the “Convention on AI for the EU” (Convention). The Convention seeks to ensure that all activities related to AI systems align with human rights, democracy, and the rule of law. It will enable the EU to be part of a unified approach to AI regulation globally and facilitate cooperation with other Council of Europe members and third countries.
Impact: The Convention will be implemented in the EU exclusively through the EU AI Regulation. It excludes national security and certain R&D activities but mandates safeguards like transparency, accountability, privacy, and access to remedies for rights violations. It sets a global precedent for binding international AI governance and complements existing human rights instruments. Read our Commercially Connected Shorts for more information.
EU: Commission launches AI tools for researchers and industry
On June 24, 2025, the European Commission (EC) launched new features on its AI-on-Demand platform. The platform now includes an AI marketplace, a low-code AI development tool, and secure solutions for generative AI and large language models.
Impact: This initiative aims to position Europe at the forefront of AI innovation by providing trustworthy AI tools and solutions, boosting the EU’s capabilities as outlined in the AI Continent Action Plan. This development is expected to accelerate secure and scalable AI adoption across European industries and the public sector.
EU: General-Purpose AI Code of Practice published
On July 10, 2025, the European Commission (EC) published the final version of the General-Purpose AI Code of Practice (AI Code). It must be endorsed by both the EC and the member states. The AI Code is a voluntary tool to help providers comply with the AI Act rules on general-purpose AI (GPAI).
Developed by 13 experts and with input from over 1,000 stakeholders, the AI Code covers transparency, copyright, and safety and security. The first two chapters apply to all GPAI model providers. They include tools such as a Model Documentation Form and practical copyright compliance solutions. The third chapter outlines risk mitigation measures for providers of models presenting systemic risks. The AI Code is complemented by the EC guidelines for providers of GPAI models.
Impact: The AI Code supports a more predictable and innovation-friendly environment for AI development and deployment in the EU. By signing the AI Code, providers can demonstrate compliance more easily. They will benefit from reduced administrative burden through standardized documentation and risk management tools. Greater legal certainty is also provided, especially for GPAI providers. Read our briefing for more information on key obligations that came into effect on August 2, 2025, for GPAI models.
EU: Guidelines for providers of general-purpose AI models published
On July 18, 2025, the European Commission (EC) published guidelines for providers of general-purpose AI (GPAI) models under the AI Act. These clarify obligations such as transparency, copyright compliance, and systemic risk mitigation.
The guidelines focus on four key topics: GPAI models, providers of GPAI models, obligations exemptions, and enforcement. GPAI models are defined by high computational training thresholds and broad capabilities. The guidelines explain when an actor becomes a provider, outlines exemptions for open-source models, and details enforcement expectations. These guidelines complement the General-Purpose AI Code of Practice and aim to ensure legal clarity and responsible AI development in the EU.
Impact: Businesses developing or integrating GPAI models must prepare for the new compliance obligations that came into effect on August 2, 2025. Those with systemic risk must notify the EC AI Office. The EC will support early compliance, with enforcement powers starting in August 2026. Models already on the market must comply by August 2027.
EU: European Commission publishes GPAI Training Data Summary Template
On July 24, 2025, the European Commission (EC) released the template for General-Purpose AI (GPAI) model providers to summarize training data. Under Article 53(1)(d) of the AI Act, using the template is mandatory.
The template promotes transparency and supports compliance with the AI Act’s data disclosure requirements. It requires disclosure of high-level information, such as the top 10% of domain names scraped, while allowing providers to withhold technically detailed or commercially sensitive data.
It also offers a standardized format to list main data collections and explain other sources used in model training. This initiative helps copyright holders and other stakeholders exercise their rights under EU law. The template complements the recently published AI Code of Practice and guidelines on GPAI rules.
Impact: Businesses placing GPAI models on the EU market must now publish a training data summary using the official template. This requirement also applies to providers of models released under free and open-source licenses. Businesses must describe their main data sources and explain other content used, without revealing trade secrets. Failure to comply may lead to enforcement under the AI Act, with fines of up to 3% of global turnover or €15m. The EC may revise the template based on technological and market developments.
EU: European Commission allegedly withdraws AI liability and patents proposals
On July 31, 2025, the European Commission (EC) allegedly confirmed the withdrawal of two legislative proposals: the AI Liability Directive and the Standard Essential Patents (SEP) Regulation. The AI Liability Directive aimed to harmonize consumer redress for harm caused by AI products and services across the EU. The SEP Regulation targeted licensing frictions in key industries like automotive and smart energy.
The EC cited no foreseeable agreement and simplification goals as reasons for withdrawal, despite resistance from lawmakers and member states. Some MEPs criticized the lack of transparency and initiated legal action to access related documents.
Impact: The withdrawal of these proposals may create uncertainty for businesses operating with AI and patented technologies. Without harmonized liability rules, companies may face fragmented national laws, with potentially increased compliance costs. The SEP Regulation’s withdrawal eases pressure on patent holders but may prolong licensing disputes.
UK: Regulators guide firms on responsible AI use
On June 2, 2025, the Financial Conduct Authority (FCA) and the Information Commissioner’s Office (ICO) published a blog post guiding firms on responsible AI use.
The FCA and ICO are collaborating to help firms use AI responsibly while protecting consumers and fostering innovation. The blog emphasizes that regulation should act as a bridge, not a barrier, enabling firms to invest and grow responsibly. It also outlines initiatives to clarify regulatory expectations. It addresses data sharing, liability, and controllership issues in the AI supply chain.
Impact: The FCA and ICO will refine regulations by developing a statutory code of practice for AI and automated decision-making. They will engage with industry stakeholders, including smaller firms, to provide clearer compliance examples and reduce uncertainty.
UK: Compliant use of AI in advertising — the UK’s stance
On May 29, 2025, the Committee of Advertising Practice (CAP) confirmed the UK requirements on the use of AI in advertising and marketing. Its key messages (which are a useful reminder of the UK’s positioning) include:
- the use of AI in advertising/marketing needs to be compliant with the CAP/BCAP codes — these codes apply regardless of how content is developed and marketed; the CAP has not developed AI-specific rules on this basis
- there is no blanket legal requirement in the UK to disclose the use of AI in ads (unlike in other jurisdictions); however, global brands will need to position themselves for compliance with legislation elsewhere (such as the EU AI Act)
- a reminder of the 12 guiding principles on GenAI in advertising that are recommended by the advertising industry (a key one being to ensure the use of AI is transparent in ads)
CAP will continually monitor its approach to ensure it matches the developments in AI.
Impact: This steer presents a useful checklist for businesses in devising and adapting their marketing campaigns.
UK: Ofcom’s updates on online safety and AI
On June 6, 2025, Ofcom published its strategic approach to AI for 2025/2026. In this, Ofcom set out how they are supporting the safe innovation and use of AI across the sectors they regulate, and in streamlining the way it works.
Impact: Ofcom’s regulation is technology neutral (meaning regulated companies are essentially free to deploy AI as they see fit, without needing permission). Ofcom has committed to supporting AI innovation through:
- the creation of safe spaces to support technology
- providing large data to train and develop AI models
- collaboration in the Digital Regulation Cooperation Forum (alongside the CMA [competition], ICO [data], and FCA [finance] regulators)
They will also mitigate the risks of AI deployment through the Online Safety Act and use AI to generate efficiencies in-house (taking a safety-first approach).
UK: AI tools set to transform UK audit practices
On June 26, 2025, the Financial Reporting Council (FRC) published its first guidance on the use of AI in audit. It also published a thematic review of the six largest firms’ processes to certify new technology used in audits.
As AI tools continue to be used in audit processes, this guidance explains how to implement a hypothetical AI tool. It offers insights into FRC documentation requirements, all designed to support innovation across the audit profession.
Developed with the FRC’s Technology Working Group and audit experts, the guidance helps auditors and central teams use AI tools effectively. It also informs third-party tech providers about regulatory expectations for their audit clients.
The thematic review shares insights on how the firms certify automated tools for audit use. It highlights the processes and controls involved, offering examples of good practice.
Impact: The guidance is intended to encourage responsible innovation as AI becomes increasingly relevant in the audit profession. Businesses considering AI adoption should evaluate whether their data quality and audit requirements support its integration and carefully select tools that are explainable and compatible with existing methodologies. When using third-party tools, it may be beneficial to seek independent assurance and be aware of any limitations in documentation. Collaboration between technology and methodology teams is encouraged to help ensure that AI tools are appropriately and effectively embedded into audit processes.
UK: How can tech firms help users spot deepfakes?
On July 11, 2025, Ofcom published a discussion paper exploring the different tools and techniques used to identify deepfakes. Deepfakes are AI-generated videos, images, and audio content that are deliberately created to look real.
The paper highlights four key attribution tools: watermarking, provenance metadata, AI labels, and context annotations. These tools aim to show how content was created and whether AI was involved. The paper looks in detail at how each measure works and assesses its strengths and weaknesses, and considers what it would take to deploy these tools successfully.
Impact: Ofcom’s research reveals findings that should guide future action by industry, government, and researchers in this area:
- evidence shows attribution tools help users think more critically when properly tested
- users shouldn’t bear full responsibility for spotting deepfakes
- platforms must support users by sharing clear, balanced AI information
- tools should explain how AI shaped content, not just flag its presence
- some tools, like watermarks, are vulnerable to simple edits and removal
- standardizing attribution tools could improve their effectiveness and adoption
- rapid tech changes mean broad claims about these tools are risky
- combining attribution with AI classifiers and reporting systems tackles deepfakes more effectively
UK: Compute Roadmap — advancing AI in the UK
On July 17, 2025, the government unveiled its new Compute Roadmap, a ten-point plan to deliver on the government’s commitment to increase the compute infrastructure within the UK. This includes:
- launching and expanding the capacity of the AI Research Resource, “a suite of advanced supercomputers that provides AI-specialized compute capacity to researchers, academia, and industry”
- setting up a network of National Supercomputing Centres (the first being in Edinburgh)
- establishing the AI Sovereignty Unit to focus on developing the UK’s domestic AI capabilities and compute ecosystem
- investing in Wales and Scotland with AI Growth Zones to aid data center development powered by responsible energy sources and support R&D
Impact: The Compute Roadmap is likely to have a big impact on clients by making advanced AI tools and infrastructure more accessible, which may mean faster innovation; better data protection; and more efficient services in areas like healthcare, finance, and education. It also brings new legal and regulatory challenges regarding how data is used and managed, so businesses are encouraged to stay on top of these changes and adjust how they work to stay compliant and competitive.
UK: Government report on the impact of social media algorithms and generative AI (genAI)
On July 11, 2025, the Science, Innovation and Technology Committee published a report on its inquiry into the spread of misinformation and the sufficiency of the online safety regime. The current online safety regulation does not go far enough, and the report urges the government to regulate further on GenAI and the spread of misinformation via social media companies by focusing on five “fundamental principles”:
- public safety — GenAI content needs clear labeling and fact-checked misinformation needs demoting by platforms; further work between government and platforms is required
- free and safe expression — a proportionate approach is necessary
- responsibility for content — platforms and advertisers are currently “unable or unwilling to address the monetization of false and harmful content”
- control over content and data — users need the power to control their information feeds (they also need to be liable for what they post and take responsibility)
- technological transparency — further work is required to uncover how algorithms operate and other measures; this is one of the key pieces of information needed to further regulate online harms
We expect the government to respond to various recommendations within two months, including:
- a greater role for Ofcom in holding platforms accountable for misinformation
- social media platforms to embed tools in their systems to deprioritize and address misinformation posts and give users control to reset their data-recommended algorithms
- minimum standards for platforms to comply with in addressing online content — risk assessments and reporting will form part of this together with tracking the source of misinformation
- regulation of “small but risky” platforms
Impact: In light of this report, businesses operating in the tech, media, advertising, and online service sectors are encouraged to:
- review their content moderation and crisis protocols
- review their policies for references to user well-being in light of harmful content and empower users to control their content
- to the extent practicable, increase transparency when engaging with public bodies
- understand their advertising supply chain
US: Department of Commerce renames and reforms AI Safety Institute
On June 4, 2025, the Department of Commerce announced the reform of the AI Safety Institute. It is now the Center for AI Standards and Innovation (CAISI). The change, directed by the US administration, aims to enhance US economic and national security.
CAISI will:
- collaborate with the National Institute of Standards and Technology (NIST) to develop guides, best practices, and voluntary standards for AI systems
- establish agreements with private-sector AI developers
- evaluate AI capabilities posing national security risks
- assess foreign and adversary AI systems
Impact: CAISI aims to protect American technologies from excessive foreign regulation. It will regularly collaborate with other organizations within NIST, as well as other bureaus within the Department of Commerce.
US: Senate decides not to impose AI law moratorium
On July 1, 2025, the US Senate removed a proposed five-year ban on the enforcement of AI legislation from the One Big Beautiful Bill Act. The proposed ban would have hampered states’ ability to impose new regulations on a wide array of issues related to the development and deployment of AI or to enforce existing laws that burdened AI innovation and adoption.
Impact: The legislation, ultimately signed by President Trump on July 4, 2025, leaves states with the authority and opportunity to continue regulating AI. Despite the failure to legislate a federal moratorium that preempts state efforts on this issue, the administration included in its AI Action Plan (discussed further below) similar concepts meant to contain states from imposing restrictive or burdensome regulations on AI.
US: White House issues AI Action Plan and executive orders
On July 23, 2025, the White House published its AI Action Plan and President Trump issued three AI-related executive orders (EOs). The AI Action Plan outlines the key priorities of the US administration’s AI policy agenda. It emphasizes rapid innovation, deregulation, infrastructure expansion, and international leadership.
The three EOs cover:
Impact: Businesses should consider:
- preparing for expanded opportunities in AI exports, infrastructure, and federal partnerships
- aligning their AI innovation with US priorities and expectations, especially around standards for factual accuracy and ideological neutrality in AI systems
- reviewing internal policies and governance frameworks to ensure responsible AI use
Infrastructure providers, especially those in cloud and data center services, should assess eligibility for new federal funding tied to qualifying projects. Read our briefing to obtain additional information.
US: California regulates business use of automated decision-making technology
On July 24, 2025, the California Privacy Protection Agency (CPPA) adopted new rules for automated decision-making technologies under its privacy law. These tools now face stricter transparency and compliance requirements. The rules package is focused on three key matters:
- the use of automated decision-making technology (ADMT) to make a significant decision affecting consumers, with increased transparency and new consumer and worker rights
- cybersecurity audits and risk assessments
- greater accountability through expanded reporting requirements to the CPPA
The rules also revise definitions for key terms like multifactor authentication, automated decision-making technology, privileged accounts, and sensitive personal information. They aim to clarify how these terms apply across different industries and data uses.
The rules spotlight complex areas such as employment decisions, housing, financial or lending services, and other areas where data processing will impact consumers’ access to critical life opportunities or services. Businesses, under these new regulatory requirements, will likely face significant operational challenges to meet their expanded consumer privacy obligations.
Impact: The rules are likely to create significant additional compliance burdens on covered businesses and employers while giving consumers new rights. Businesses should:
- inform consumers before using ADMT tools to process their personal data; notices must explain the tool’s purpose, logic, and consumer rights clearly
- ensure processes are in place to produce the documentation necessary to conduct risk assessments of their data processing activities and to submit timely reports to the CPPA beginning April 1, 2028
- prepare to determine whether they will need to comply with the cybersecurity audit requirements, and have in place procedures to ensure that depending on the size of the business, the company will be able to meet the first reporting deadline (as early as April 1, 2028, for those with annual revenue greater than $100m)
Read our briefing for further information. The rulemaking package will be submitted to the California Office of Administrative Law (OAL). The OAL has 30 business days to review and approve the regulations.
US: Colorado to convene special legislative session to consider implementation changes to pioneering AI law
The Governor of Colorado, Jared Polis, announced the Colorado General Assembly will have a special legislative session beginning August 21 that will in part address the fiscal and implementation impacts from the Colorado Artificial Intelligence Act. Legislators earlier this year sought to codify recommendations to push back the implementation date for the law, but those efforts failed. Governor Polis has consistently indicated that the law the legislature passed and that he ultimately signed would need to be improved down the line, with technology companies and industry groups consistently calling for revisions to the law that would promote innovation.
Impact: Colorado’s AI law imposes various obligations on both developers and deployers of certain AI systems. Any extension of the implementation timeline would likely afford businesses greater flexibility to ensure they have the necessary governance practices in place to ensure compliance with the law’s disclosure and reporting requirements.
US: National Institute of Standards and Technology is developing resources to meet demand for additional AI guidance
Under the AI Action Plan, the National Institute of Standards and Technology (NIST) will continue to play a significant role in fulfilling almost two dozen initiatives. The standards-setting organization is already in the process of developing a timeline for its activities to deliver on its assignments and meet the administration’s expectations. In addition to those efforts, NIST is seeking public comment on a concept paper and proposed action plan for developing a series of controls for securing AI systems in addition to its foundational guidelines for managing the risks from AI.
Impact: NIST has played a pivotal role in shaping the AI governance practices of businesses through its AI Risk Management Framework. While NIST’s role could have shifted dramatically, the administration has continued to prioritize the organization’s leadership in crafting foundational elements of US AI policy and to empower its efforts to develop technical standards and metrics for evaluating AI performance.
US: Lawmakers in Texas enact the Texas Responsible Artificial Intelligence Governance Act
Texas Governor Greg Abbott signed into law legislation that establishes a framework for regulating the use of AI by individuals, businesses, and governmental entities in the state. The law lays out six prohibited uses of AI and outlines how the specific restrictions apply to different actors.
The banned uses include the development or deployment of AI systems to (1) manipulate human behavior, (2) calculate a social score, (3) violate constitutional protections, (4) capture biometric data without consent, (5) produce sexually explicit content and child pornography, and (6) unlawfully discriminate. The law focuses on the intentional use of AI in carrying on those restricted activities as the basis for establishing liability and protects developers from responsibility for misuse by third parties.
The law also includes requirements for government agencies to provide reasonable notice to consumers when those consumers are interacting with an AI system deployed by the agency. The Texas Responsible Artificial Intelligence Governance Act (TRAIGA) establishes a regulatory “sandbox” program to give AI developers time to test and develop their technologies without needing to comply with specific regulatory obligations. The program is meant to incentivize companies to develop their AI systems in the state without the looming risk of government enforcement or regulatory limitations constraining their innovation. Texas is also going to establish an Artificial Intelligence Council that will serve as a body of experts to provide insights, recommendations, and training to ensure the effective oversight of the sandbox program and continued development of Texas’s AI laws.
Impact: Developers and deployers of AI systems will have until January 1, 2026, to come into compliance with TRAIGA. For many businesses, this will involve operationalizing a comprehensive AI risk-management program to gain further protections from liability under the law.
For additional information, please read our briefing on the new law.
Harmonized AI standards to reduce fragmented global rules
On July 11, 2025, the International Chamber of Commerce (ICC) published a policy paper urging harmonized, market-driven AI standards. Small and medium-sized enterprises (SMEs) face major challenges managing complex, country-specific AI rules. They often lack the resources to handle overlapping legal requirements. The paper highlights how divergent AI regulations across countries can lead to fragmented global markets and increased business costs.
The paper calls for greater coordination on the development of international, market-driven AI standards. The aim is to bridge legal differences, reduce compliance burdens, improve market access, and enhance cross-border innovation.
Impact: The ICC has made the following recommendations for policymakers to make AI standards work globally:
- promote strategic alignment in AI standards development to reflect market needs and avoid duplication
- ensure domestic and local expert participation in shaping market-driven standards
- prioritize global, industry-driven standards over national or regional-only approaches
- champion multistakeholder collaboration through transparent, inclusive processes
- leverage existing standards in regulation to streamline compliance and build trust
- use standards in public procurement to support adoption and open markets to SMEs
- support company participation with funding, incentives, and training
- enhance awareness and education to build capacity for implementing AI standards
Co-authored by Jon Botham (Knowledge)
Further reading