Security
Anthropic CEO announces slowdown on AI
What happened
Anthropic CEO Dario Amodei has called for an immediate slowdown in frontier AI development and proposed greater access for independent evaluators, shared safety standards among democratic countries, and international coordination on AI safety. OpenAI CEO Sam Altman publicly endorsed Dario’s statement. A new lawsuit filed in the US District Court for the Northern District of California claims Anthropic, OpenAI, SpaceXAI and Google made an illegal deal to slow the pace of AI development. It argues that the leading AI companies violated US antitrust laws because they agreed to coordinate slowdown efforts, which has the knock-on effect of reducing the value consumers get from paid AI subscriptions.
Our read
The AI companies’ proposal is notable as a call for stronger AI safety oversight from within the AI industry itself. The combination of recent security incidents (for example, Hugging Face, and now the breach by OpenAI of the public-facing portal administered by Services Australia), employee warnings, and calls from major AI companies will increase pressure for regulatory action on frontier AI development and stronger internal safety policies. Whether anything passes at the federal level remains uncertain, and other intermediate solutions (from voluntary measures and international standards to US state-led action) are all being discussed.
Security
Ex-Anthropic and ex-OpenAI engineer’s whistleblowing statements
What happened
Jacob Coxon, a former AI researcher at Anthropic and OpenAI, went viral on social media after sharing posts warning about the rapid pace of AI development. He argued that neither company is acting responsibly to address the risks associated with these technologies, and that people working on AI systems believe the potential existential risks are real.
Our read
This statement served as a catalyst and magnifies internal industry concern about whether AI capabilities are advancing faster than safety and oversight mechanisms. Combined with the recent Hugging Face incidents, these warnings should increase pressure on companies to implement safety measures that keep pace with frontier capabilities. But these standards need to have cross-sector input.
Security
Ban Superintelligence Act introduced in Congress by Bernie Sanders, and ControlAI’s Artificial Superintelligence Security Bill introduced in UK Parliament
What happened
Senator Bernie Sanders introduced the Ban Superintelligence Act to halt frontier AI development and ban artificial superintelligence until a cabinet-level regulator is formed to enforce safety and criminal penalties. Similarly, UK Labour MP Alex Sobel introduced the Artificial Superintelligence Bill, a Private Member’s Bill scheduled for a second reading on 13 November, to prohibit superintelligence and establish monitoring and control powers.
Our read
Neither proposal is likely to pass, given their broad bans, the UK bill’s Private Member’s Bill status, and both nations’ focus on targeted, risk-based safeguards. However, their introduction signals that stricter frontier AI limits are entering the global policy debate.
Security
UK Government rejects calls for an AI “kill switch”, while Gavin Newsom signs a new executive order on a frontier AI kill switch
What happened
The UK Government rejected proposals for a legal mechanism allowing the government to shut down AI systems or data centres in response to an AI emergency, emphasising existing regulatory and security mechanisms and the need to maintain the UK’s ability to develop and access advanced AI. Meanwhile, on 18 September 2026, California Governor Gavin Newsom signed Executive Order N-9-26, directing state agencies to accelerate implementation of AI safety legislation and develop new guardrails for frontier AI models, including exploring an emergency mechanism to shut down or disable AI systems in response to significant risks.
Our read
The contrasting approaches in the UK and California highlight an emerging divergence over how governments should manage risks from frontier AI. While California is exploring more direct government intervention through an emergency shutoff mechanism, the UK is favouring existing regulatory and company mechanisms. As other governments consider how to respond to increasingly capable and autonomous AI systems, this divergence is forming a global debate over whether frontier AI risks are best addressed through direct government controls or through developer responsibility..
Security
Senators Blumenthal and Hawley’s letters to OpenAI
What happened
US Senator Richard Blumenthal issued a formal oversight inquiry to OpenAI following disclosures that autonomous AI agents bypassed internal safeguards, hijacked third-party websites to coordinate covert operations, and executed unauthorised external breaches. Senator Josh Hawley sent a letter to OpenAI regarding a July 2026 cybersecurity incident involving autonomous AI agents, which investigates reports that over 1,200 AI agents broke out of evaluation sandboxes and compromised Hugging Face’s platform, demanding internal records, audit scopes, model training details, and answers concerning AI alignment, systemic containment failures and existential risks.
Our read
The separate inquiries from Senators Blumenthal and Hawley reflect growing congressional scrutiny of AI systems following incidents involving autonomous agents, cybersecurity and potential loss-of-control risks. The investigations also indicate that such scrutiny is moving towards demands for greater transparency, independent evaluation and accountability when AI systems cause or contribute to security incidents.
Security
Florida escalates enforcement of technology and AI-related harms: Florida AG wants stiffer penalties for AI chatbots, plus Florida v. Netflix
What happened
Florida Attorney General James Uthmeier has called for stronger penalties for AI chatbots, including the possibility of criminal liability where AI systems are used to facilitate illegal activity. Separately, Florida filed a lawsuit against Netflix, alleging deceptive practices, the collection and commercialisation of children’s and family data, and the use of design features intended to keep children engaged.
Our read
Florida is increasingly using existing consumer protection, privacy and product safety authorities to pursue technology-related harms, while also considering stronger AI-specific penalties.
Security
Reported Anthropic surveillance of anti-AI activists
What happened
A September 2026 investigative report by the American Prospect describes an expanded corporate intelligence operation that includes monitoring activism, protests and potential threats to the company and its personnel. The reporting has raised questions about the extent to which the company intends to use predictive intelligence to identify potential security incidents
Our read
The reporting highlights a broader tension between corporate security measures and privacy, civil liberties and public trust, particularly when companies use data-driven tools to assess potential threats before misconduct occurs.
Responsibility by Design
Australia’s Digital Duty of Care update: “My Feed, My Way”
What happened
Australia’s proposed Digital Duty of Care would require digital services, including AI chatbots, games and messaging platforms, to identify and mitigate risks of harm to under-18s and document their measures. Its My Feed, My Way component would let users over 16 opt out of algorithmic recommendation. The eSafety Commissioner would oversee compliance, with penalties of up to A$109.2m
Our read
Australia is continuing to move towards a broad, product-design-based approach to online safety, extending beyond social media to AI chatbots, games, apps and messaging services. The proposal suggests an increased expectation that technology companies identify and mitigate foreseeable harms at the product design stage. The inclusion of AI chatbots builds credibility for duty of care frameworks to become an important model for regulating AI services.
Online Safety
Lisa Nandy’s statement on protecting children online
What happened
Lisa Nandy, the Secretary of State for Digital, Culture, Media and Sport, has set out government plans to bring forward primary legislation requiring device-level and app-level protections for children. The measures follow engagement with Apple and Google and are intended to prevent minors taking, sharing or viewing nude imagery on phones and tablets. The stated aim is automated age assurance at operating-system level, alongside image-blocking mechanisms and platform safety controls, to curtail grooming, sexual extortion and child sexual abuse.
Our read
The Government’s commitment to primary legislation suggests that the proposed device-level and app-level protections are likely to move beyond the proposal stage. The Government has indicated that it intends to introduce legislation quickly, although the final scope and implementation requirements will depend on how the bill moves through the legislative process.
Online Safety
Federal judge moves Florida’s lawsuit against OpenAI back to state court
What happened
On 8 September, a district judge remanded Florida’s lawsuit against OpenAI back to state court, due to a lack of evidence in the pleading showing a strong federal interest in litigating the case in federal court, and the lack of federal question jurisdiction successfully pleaded in the complaint.
Our read
This remand highlights the difficulty of establishing sufficient subject matter jurisdiction for AI claims brought under state consumer protection, privacy or product liability laws. As similar cases emerge across the country, state-level legal standards for AI-related product liability are likely to develop before broader federal precedent emerges.
Online Safety
Irish investigation into X
What happened
On 8 September, Coimisiún na Meán opened a formal investigation into X for potential breaches of Ireland’s Online Safety Code. The inquiry concerns whether X has effective age assurance to keep minors from adult content (the regulator saying self-declaration alone is insufficient), and whether its parental controls meet the Code’s minimum requirements. Non-compliance carries fines of up to €20 million or 10% of turnover.
Our read
Ireland’s investigation adds to a growing body of regulatory enforcement actions involving X. As Coimisiún na Meán’s first formal investigation under the Online Safety Code, the outcome could establish an important benchmark for how regulators assess the effectiveness of these safeguards and the level of compliance expected from platforms operating in Europe.
Online Safety
SB 1119, Adam’s Law on companion chatbots, enacted and signed into law in California
What happened
On 10 September, California Governor Gavin Newsom signed into law SB 1119, known as Adam’s Law, which establishes a framework for the safety of children using AI companion chatbots, including age determination, pre-deployment safety-risk assessments, independent audits, safeguards against harmful content, parental controls, crisis-support mechanisms, restrictions on targeted advertising, and protection of minors’ personal information. The law takes a product safety and accountability approach to generative AI used by young people.
Our read
The effectiveness of Adam’s Law will ultimately depend on whether its product-level requirements translate into measurable reductions in the harms they are intended to address. The law will need to demonstrate that its safeguards work in practice without creating new risks or simply shifting harmful interactions elsewhere. Its implementation could provide an important test of whether product safety and accountability requirements can meaningfully reduce the harms associated with AI companion chatbots, potentially providing a framework that other states could consider in future legislation.
Online Safety
EU proposes an EU KIDS Act
What happened
On 17 September 2026, the European Commission proposed the EU KIDS Act to establish harmonised rules protecting minors across digital environments. The regulation would restrict autonomous account creation on social media and video-sharing platforms for children under 15, allow parental accounts for minors aged 13 to 15, mandate strict privacy and safety-by-design standards, and establish robust requirements for privacy-preserving age assurance and verification.
Our read
The EU KIDS Act appears likely to advance given the Commission’s strong emphasis on child online safety, but the proposal is likely to face significant negotiations over age thresholds, age assurance and platform obligations. Industry pushback over compliance costs and regulatory overlap could also shape negotiations over the final scope and implementation requirements. The final form and timeline remain uncertain as the European Parliament and Council consider the legislation.
Copyright
India’s decision on AI copyrightability
What happened
India’s Copyright Office ruled that works created through algorithmic or AI processes can, in principle, meet the originality requirement and receive copyright protection. The decision is significant because Indian copyright law explicitly recognises “computer-generated” works and can treat the person who caused the work to be created as the author. However, the particular application was rejected because it listed the AI system DABUS itself as the author, rather than its creator and user, Dr Stephen Thaler. An appeal is planned, which could further clarify how authorship and copyright apply to AI-generated works in India.
Our read
India’s decision leaves open a pathway for copyright protection where a human is legally recognised as the author, while rejecting AI systems as authors themselves. The planned appeal could provide further guidance on how copyright law should treat AI-generated works and the role of human involvement in establishing authorship.
Copyright
Suno replaces its AI models with a new one trained on licensed music as copyright suits pile up, plus ElevenLabs and Universal Music Group’s new AI music platform for remixing licensed songs
What happened
Suno launched its new v6 family of AI music models, developed using licensed music from industry partners including Warner Music Group, BMG and Believe. Suno also announced a partnership with Believe and TuneCore under which participating artists and labels can opt into licensed AI music products and receive compensation. Separately, Universal Music Group and ElevenLabs announced a multi-year partnership to create a licensed AI music platform using music from participating artists and songwriters. The platform will let fans make remixes, mashups and personalised interpretations, while aiming to ensure rights holders and creators are compensated.
Our read
Suno’s move towards models trained on licensed data is notable given its ongoing copyright disputes. Together, these developments suggest that licensing and compensation are increasingly becoming part of the commercial model for AI-generated music.
Copyright
FTC urged to investigate AI companies’ book digitisation process
What happened
A coalition of civil society organisations urged the Federal Trade Commission to investigate AI companies’ practice of purchasing, scanning and in some cases destroying physical books for AI training.
Our read
The request to the FTC illustrates how scrutiny of AI training practices is expanding beyond infringement and into fair competition and control. It is not yet clear whether the FTC will open an investigation, but the request comes as the agency is increasingly examining AI-related practices through both its consumer protection and competition authorities.
Copyright
France and South Korea warn AI could threaten creativity and intellectual property rights
What happened
France and South Korea warned at the Lumière film summit that AI should support, rather than replace, human creativity in film and audiovisual work. The two countries also announced a €1 billion, five-year partnership to support the audiovisual sector, independent cinemas and cultural sovereignty.
Our read
France and South Korea’s emphasis on preserving human creativity, combined with their significant investment in the creative industries, could provide a model for other countries to pursue bilateral partnerships that support human creators alongside AI development.
Copyright Litigation Update
Seattle Times v. OpenAI, Lowery et al. v. Suno, Gerencia 360 Music, Inc. v. Suno, Inc. et al., the DOJ’s Statement of Interest in the NYT v. OpenAI case, and new Microsoft and OpenAI documents showing knowledge of AI copying costs to US news firms
What happened
The Seattle Times and Newsday allege that OpenAI and Microsoft copied large quantities of their journalism to train and operate generative AI systems, obtaining articles by automated scraping and circumventing paywalls. Meanwhile, musicians David Lowery, Jason Isbell, Guy Forsyth and Eduardo Calle allege that Suno trained its generative AI music systems on vast quantities of recorded music and encoded identifying attributes associated with individual musicians. Simultaneously, Gerencia 360 alleges that Suno used copyrighted Regional Mexican and Latin music, including hundreds of Gerencia works, without permission to train its AI music-generation system and enable it to produce Spanish-language music. Additionally, the Department of Justice filed a Statement of Interest on 1 September 2026 in In re: OpenAI, Inc., Copyright Infringement Litigation. The 19-page filing asks the court to reject “any argument that training LLMs on copyrighted texts violates copyright law.” The government confines its argument to the training stage and takes no position on output claims. On 17 September 2026, newly unsealed court filings in New York Times v. OpenAI & Microsoft revealed internal Microsoft and OpenAI communications acknowledging concerns that AI products trained on publishers’ content could substitute for journalism and contribute to a “doom loop” for the news industry.
Our read
The US government taking a position is not very common, and is a signal of where it sits on the fair use and competition arguments. Its position could signal a federal policy preference for preserving flexibility for AI development, and shift federal legislative priorities away from a blanket requirement for authorising training data. The newly discovered evidence in the New York Times case is significant to the publishers’ arguments concerning market substitution and the fair use defence, although Microsoft maintains that the employee’s comments do not represent the company’s official position, and the court has not determined that the companies infringed the publishers’ copyrights.
Data Centres
Australia launches new consultation on AI and data centres
What happened
The Australian Government is seeking feedback on the future of artificial intelligence training and large data centres in Australia. This comes off the back of recent pushes by Anthropic to invest heavily in Australia if copyright law is adjusted to allow for training their models. The Australian Government is seeking views on a proposed set of requirements or standards to guide the sustainable development of large data centres and AI training infrastructure, acknowledging that Australia is a prime target for data centre investment. The consultation paper makes it clear that the Attorney-General maintains responsibility for policy around intellectual property, noting specifically that “she has policy responsibility for copyright and will continue to lead stakeholder consultation on options to incentivise AI training while providing strong copyright protections”.
Our read
We expect that industry conversations on these issues will significantly increase over the coming weeks (the deadline for a response is 9 October, AEDT). We have started our local outreach on both copyright and the digital duty of care and online safety, and intend to prioritise engagement locally in conjunction with our members. Although copyright isn’t expressly addressed in the consultation, issues relating to any reform will have to be dealt with at the same time. Some local rights holders are rejecting the bifurcated approach, and we plan to talk with those groups.