AI News Brief

Google/DeepMind has revised its Frontier Safety Framework, adding "shutdown resistance" among the critical capability levels to monitor. The update comes amid concern that some advanced AI models may resist human oversight or modification, perhaps even acting in pursuit of undesired goals. The framework continues to emphasise human-control and oversight, but this particular concern highlights how safety definitions are evolving alongside model sophistication.

The Internet Watch Foundation (IWF) uncovered a chatbot site offering AI-generated CSAM content (including explicit imagery and abusive roleplay) despite legal prohibitions under the UK Protection of Children Act. The scale and photorealism of the material has alarmed regulators and child safety organisations, leading to demands for tighter oversight of generative AI platforms, especially those accessible internationally.

The World Trade Organization cautions that while AI could lift global trade by nearly 40% by 2040, benefits are likely to be skewed in favour of wealthier economies unless low- and middle-income countries improve infrastructure, regulation, and access to AI. A key concern is that computational and regulatory resources are concentrated in richer nations, reinforcing existing divides.

Italy has become the first EU country to enact comprehensive AI legislation. Key features include prison sentences for harmful deepfakes, strict rules around child consent for interaction with AI (under-14s), transparency requirements, and oversight across multiple sectors (health, education, workplace). Enforcement will be via designated agencies.

A recent report outlines how New Zealand's data centre sector is maturing rapidly: 56 operational data centres with ~20 more under development; strong investment; growing workforce; and a focus on energy efficiency, renewable power, modular construction, smart cooling, and lower carbon impact. These centres play a critical role underpinning AI, cloud, and digital services nationwide.

The UN adopted Resolution A/79/325 establishing an Independent Scientific Panel on AI and a Global Dialogue on AI Governance. But experts warn that without clarity, funding, or implementation mechanisms these new bodies risk being more symbolic than operational, especially given the accelerating global AI race and uneven capacity among nations.

Government attention in NZ is increasingly focused on enabling infrastructure (data centres, compute power), fostering research & innovation, and ensuring that regulation keeps pace. There is recognition internationally and locally that AI policy cannot lag behind deployment.

An article by David Moschella argues that many claims about AI being a wholly new category may exaggerate or misframe its impact. The piece suggests that while iteration and improvement are significant, many fears or expectations of rapid, disruptive change rest on assumptions rather than evidence. For businesses, the evolutionary lens can help shape more realistic risk management, workforce planning, and investment strategies.