UTS hosts global AI Policy Roundtable, shares key insights

Today we released an open letter to political leaders calling for a bipartisan effort to regulate AI in Aotearoa. AI is already transforming healthcare, education, and work. But right now, New Zealand has no dedicated AI law. That leaves us exposed to big risks, while also creating uncertainty for businesses that want to innovate responsibly. That’s why 20+ AI experts (myself included) are calling for a risk-based regulatory approach: high-risk AI systems should face stronger guardrails, while low-risk, everyday uses of AI can continue with minimal red tape. This way, we can protect people without stifling innovation. You can read the full letter (and add your support) here: regulateai.nz #AI #Aotearoa #Governance #Regulation #TechForGood With: Christopher McGavin, Cassandra Mudgway, Joshua Yuvaraj, Ali Knott, Michael Daubs, PhD, Olivia J. Erdelyi, Allyn Robins, Anthony Robins, Andrew Chen, Ethan Plaut, Caleb Moses, Peter Thompson, Ian W., Kevin Shedlock, Hui Ma, Heitor Murilo Gomes, Emily O'Riordan, Lee Timutimu, Frith Tweedie, Olivier Jutel, aimee whitcroft, Marcus Frean, Grant Dick, Veronica Liesaputra, Brendan McCane, Stephen Cranefield

California’s SB 53 could set new standards for AI safety, but its approval is still uncertain. Supporters say it positions the state as a leader in responsible AI, while critics warn it may burden startups and conflict with federal efforts. The bill raises a key question: how do we balance innovation with accountability? Businesses should be preparing now for what may come next. Learn more: https://lnkd.in/gu3cnW9G #AIGovernance #ResponsibleAI #AICompliance #California

In our recent commentary "The Silent Collapse of AI Ethics Safeguards" (https://lnkd.in/dijdFw33, Norwegian version can be found here: https://lnkd.in/dfsJfUx8), Klas Pettersen and I documented and discussed how copyright and deepfake protections in major AI systems got less important almost overnight, coinciding with U.S. court rulings on "fair use" and the Trump administration’s July 2025 executive order on "woke AI" (https://lnkd.in/dAujQt6S). Now, in the New England Journal of Medicine - NEJM AI, David Blumenthal describes the same phenomenon unfolding in healthcare. The very policies we flagged as opening the floodgates for cultural and political deepfakes are also lowering barriers in clinical AI. This is risking patient safety, inclusivity and trust (https://lnkd.in/dCG3tziT). Since then, the legal picture has shifted again. Earlier this month, Anthropic agreed to pay $1.5 billion to settle a class action from authors whose books were used without permission (https://lnkd.in/dqrFgcga). Judge Alsup had initially ruled parts of the training process as "fair use", but Anthropic’s massive archive of pirated works crossed the line. This case makes clear that the "fair use" justification is not a blank check on the copyright side. First politics told companies everything goes. Then courts toyed with calling it "fair use". Now a $1.5B payout shows the bill has come due. If ethical rules can vanish whenever the legal weather changes, they were never rules, just marketing. This directly affects security, safety and trust. In high stakes and sensitive domains, institutions should think carefully about which systems they rely on. #AIethics #AIsafety #AI #FairUse #Copyright #AIregulation #HealthAI #ResponsibleAI #GenerativeAI

States Are Shaping the Future of AI At National Conference of State Legislatures (NCSL)'s recent Legislative Summit, leaders emphasized that AI should solve real problems—not be used just because it’s trendy. Over 1,000 AI bills have been introduced this year, doubling last year’s count. Despite federal pushback, states are asserting their role in AI governance. 🔍 Key takeaways: -Legislators are using AI to streamline workflows like drafting and transcription. -Training public sector workers in AI remains a major challenge. -Human oversight is critical—AI can assist but not replace decision-making. #AI #GovTech #PublicPolicy #NCSL #ArtificialIntelligence #DigitalTransformation #lawyers #law #statelegislation #statepoliticis #federal #legioslatures https://lnkd.in/eFY8USqp

California may soon become the testing ground for one of the first state-level AI safety frameworks. Governor Newsom suggested this week that legislation requiring disclosure of safety protocols, incident reporting, and whistleblower protections “strikes the right balance” between addressing risks and supporting innovation. Regardless of how the bill unfolds, this moment underscores an important reality: -AI is maturing fast enough that State led oversight is now on the table. -States are racing to define what “responsible AI” means in practice. -The choices made in California will likely influence the broader national and even global conversation. For those of us building and using AI, the takeaway is clear: the conversation is shifting from what AI can do to how it should be governed. That’s a signal of just how central these technologies have become to everyday life. I’m curious, do you think state-level regulation is the right way forward, or should this be handled nationally (or even globally)? https://lnkd.in/gEryrk79

Richard Sentinella, PhD and I wrote an article for the IAPP about current US AI governance at the federal level. It walks through the history and context of AI development as well as recent updates like the AI Action Plan released this summer. Check it out here → https://lnkd.in/esu7KAVu

𝘼𝒅𝙢𝒊𝙣𝒊𝙨𝒕𝙧𝒂𝙩𝒊𝙫𝒆 𝑾𝙞𝒔𝙙𝒐𝙢 AI can supercharge government—but it can’t judge. Mauricio Covarrubias reminds us that the real frontier isn’t automation, it’s human judgment in contexts where values collide: legality vs. legitimacy, equity vs. efficiency, innovation vs. inclusion. https://lnkd.in/dknZDe5u A very interesting short piece that hits the nail with AI and Public Sector, as to me. #PublicSector #AI #AdministrativeWisdom #DigitalGovernment #Ethics

Tracking disinformation or platform accountability? Let’s compare notes 🎯 We are bringing together researchers and civil society organisations who have monitored social media or worked on data access in the past two years. During this time, the EU’s AI Act and Digital Services Act (DSA) came into force, generative AI tools entered mainstream use, and over a dozen European elections faced interference. Meanwhile, platforms cut back moderation and fact-checking, leaving critical gaps for us to fill. Together, we’ll share findings and lessons from the field in our upcoming roundtable 🤝🏽 Breakout discussions will explore: 📜 The DSA and AI Act, and what they mean for accountability 🤖 Generative AI’s role in shaping political discourse and the need for audits 🌍 Foreign interference, inauthentic behaviour, and emerging threats The sessions will be co-moderated by: 🎙️ Armin Rabitsch, Election-Watch.EU 🎙️ Carlos Hernández-Echevarría, Maldita.es 🎙️ Sonia Horonziak, PhD, Instytut Spraw Publicznych These discussions will feed into a meta-analysis report to be published later this year, so don't miss this chance to have your experience represented! Register here 👉🏽 http://bit.ly/4lSfSSb Sorina M. Ognjan Denkovsķi Michael Meyer Camila Weinmann Francesca Giannaccini Duncan Allen. Stiftung Mercator GmbH

GPAI rules in, new paper out: "AI, Digital Platforms, and the New Systemic Risk," written with the wonderful Atoosa Kasirzadeh and Lilian Edwards. We contribute to an urgent and unsettled debate around the concept, and implications, of different systemic risk standards in regulation. This discussion has become especially timely with the AI Act’s GPAI rules on systemic risk now in force. The paper develops a framework for understanding systemic risk in AI and digital platforms, drawing on lessons from finance and financial regulation, climate change, and cybersecurity. We argue that systemic risks in AI cannot be reduced to the capabilities of a single frontier model. Instead, they can arise through interactions across models (incl. between multiple agents), platforms, and institutions. Some of our key takeaways: 1) Systemic risks are collective in nature, with cascading effects that go beyond individual harms. Conceptionally, they include discrimination at scale, systematic hallucinations, and environmental harms. 2) We identify four levels of AI systemic risk: single-model, multi-model, model-platform integration, and model-institution integration. 3) The DSA provides a more comprehensive articulation of systemic risk than the more recent AI Act, which ties systemic risk too narrowly to “the most advanced” GPAI models and to market impacts. A central critique concerns the AI Act’s definition of systemic risk. By linking it specifically to “high-impact capabilities” of the most advanced models and to effects measurable on the Union market, the Act may fail to capture many risks that in practice threaten societies most severely. Large-scale discrimination, systematic hallucinations, and widespread misinformation destabilize institutions and undermine fundamental rights, yet they may fall outside the current legal definition. The Code of Practice compounds this problem by explicitly listing risks such as cyber-offense or loss of control, but leaving fundamental rights harms sidelined. This narrow framing risks misclassifying models: some less “advanced” models widely adopted across contexts in fact generate systemic risks, while frontier models with limited societal reach may not. Without recognition of collective harms such as discrimination or hallucinations, the AI Act’s systemic risk regime risks focusing on abstract catastrophic scenarios while overlooking pressing, large-scale threats to democratic processes and equality. We propose reforms that broaden systemic risk assessments, strengthen coordination between regimes, and explicitly incorporate collective harms. Full paper: https://lnkd.in/eTakZbNt and https://lnkd.in/e2JNQWNG   This is an early draft - comments and critique most welcome! #AI #SystemicRisk #AIAct #DSA #AIGovernance #PlatformGovernance #FoundationModels #AIPolicy #AIEthics #DigitalPlatforms #AIGovernanceFramework #ResponsibleAI #ArtificialIntelligence

Incorporating AI in legal practice is far less revolutionary than people outside the profession think. We, as lawyers growing our skills through the implementation of artificial intelligence, aren’t discovering new proteins, eliminating entire industries, or changing the norms and standards of our systems and society. We are, however, in the words of a mentor of mine at the firm, “farmers discovering the plow.” The work of farmers doesn’t change: the crop output is only greater, the farm more efficient, and the farmer more economically desirable when a new tool is found and properly learned. Too many times, my non-attorney friends have asked me if I fear we will see a precipitous drop in junior associates joining private practice in coming years. While the market is historically and famously unpredictable, I can confidently say there is always room for more talent. The talent itself, however, will change—to keep my metaphor— from those with the strongest hands to those with the most knowledge of how to guide the plow. (And please forgive the use of em dashes, as ironic as it would be to write this post with AI, I assure you that is not the case.) At Freshfields, we continuously strive to improve our utilization of a lawyer’s tools, including AI, at the highest level. I expect recognition of this core value in our training and vision to only grow. Congratulations to all our European colleagues for recognition by the Financial Times!