Hey
Just back from Melbourne this weekend - very productive week in town catching up with the team behind RMIT's new Generative AI Lab for Education (GAILE) and a whole bunch of others across the university and beyond. Fantastic, very productive time plotting exciting new things for 2026.
Also a great reminder that Melbourne’s weather is a rollercoaster and that, even if its nominally summer there, you'll need a jacket and an alarming amount of coffee. Worth it.
Enough of that - let’s get amongst it with the headlines from the world of AI:
Governments Move on AI While Universities Debate Assessment | Governance Speed 🎯
Big AI moves from two leading Asian governments - China launched public consultation on regulating AI anthropomorphism while Singapore created a National AI Council chaired by the Prime Minister. China's proposed law (original here and commentary here) defines anthropomorphic AI, prohibits emotional manipulation, requires mental health safeguards and guardian controls for vulnerable users, establishes provider liability for harms, and explicitly states providers "should not use replacing social interaction, controlling users' psychology, or inducing addiction as design goals". An overdue move given AI’s track record in this space to date. Singapore's council will commission AI missions across key sectors and position the nation as a trusted hub deploying solutions "faster than larger countries". Both moved within weeks.
China acknowledges "AI-related human vulnerabilities" with context-specific technical measures - usage time limits, dependency warnings, emergency contact protocols for vulnerable users, prohibition on simulating deceased relatives. Singapore's PM devoted ten minutes of his Budget speech to AI, creating an inter-ministerial council to "move with speed and scale". Strong leadership from these governments - treating AI as strategic infrastructure requiring coordinated response. Lessons there for HE more broadly.
AI Companies Fund Political Mobilisation While Research Documents AI-Enabled Influence | Advocacy Gap 📢
New paper on “cyborg propaganda” marks an interesting evolution in the dis- and mis-information space. Verified humans using AI-generated scripts to create coordinated campaigns that look like spontaneous public sentiment. A central organiser sends directives, AI generates thousands of unique captions, and real users review and post. The result bypasses bot detection because accounts are genuine and content is human-approved. Kunst identifies a “massive regulatory paradox: How do you regulate coordinated inauthentic behaviour when the ‘bot’ is a real citizen exercising free speech?” The distinction between grassroots activism and automated influence is collapsing.
Meanwhile, Anthropic contributed $20 million to Public First Action, a bipartisan 501(c)(4) working to “mobilise people and politicians who understand what’s at stake in AI development”. The sum is nominal at best for Anthropic but marks AI companies entering political advocacy directly as the policy window narrows. AI companies fund political mobilisation as researchers document how AI makes grassroots advocacy indistinguishable from manufactured campaigns.
Courts Rule AI Work Lacks Professional Protection | Accountability Gap 📋
Using AI for legal documentation? Don’t. A Manhattan federal judge ruled AI-generated documents created by a defendant then forwarded to counsel don't qualify for attorney-client privilege. The reasoning: privilege protects communications made for legal advice, not materials created independently then routed to counsel. What AI changes is scale - there are reportedly widespread cases of people using AI to generate case materials, assuming forwarding them makes them privileged. This ruling says otherwise with a clear focus on what the document is and how it came to exist, not its destination. Someone using AI to organise facts may be creating discoverable material outside privileged relationship.
The WEF's new report on AI agents governance documents this gap systematically. Agents perform expert tasks - legal analysis, medical diagnosis, financial modelling - but operate outside accountability frameworks governing human professionals. When you hire a lawyer you get liability insurance, malpractice recourse, licensing. When you use AI for legal work you bear all the risk. Universities verify students can perform expert tasks while courts confirm AI-generated work lacks professional protection. We're certifying capability for a world where capability without accountability carries all the liability.
18 Months of Assessment Development Meets Single Model Release | Ground Shifts 🌊
Danny Liu documents something rarely seen in higher education: genuine sector-wide convergence. Ten Australian universities - Sydney, Melbourne, ACU, Curtin, Newcastle - spent eighteen months building coordinated two-lane assessment frameworks: secure assessments verify students can code, analyse, write independently; open assessments accept AI integration. The architecture is rigorous. The collaboration is real. Liu's documentation reveals both the achievement and the challenge it faces: frameworks designed for February 2024 AI are being deployed into February 2026 reality where the capabilities being verified are exactly what tech insiders describe handing entirely to AI.
Students told the Digital Education Council they need frameworks for using AI to build understanding, not just produce output. They want "AI auditing" capabilities - evaluating quality, detecting bias, understanding failures. Universities built frameworks to verify independent performance whilst the ground shifted beneath them. The question isn't whether institutional response was rigorous enough. It's whether any institutional response can match acceleration when 18 months of policy development becomes obsolete between design and deployment.
“AI Just Took My Job” Goes Viral | Ground Shifts 🌊
Matt Shumer's essay hit 30 million reads in 24 hours because it named what tech workers already experienced: AI just took their jobs. The startup CEO describes his new workflow - describe what you want built, walk away for four hours, return to finished work done better than he'd have done himself. GPT-5.3 Codex (released February 5th) "was instrumental in creating itself" - OpenAI's documentation confirms the recursive improvement loop researchers warned about. The AI helped build the next AI. Anthropic’s CEO predicts models "substantially smarter than almost all humans at almost all tasks" by 2026-2027. Shumer's warning to family and friends went viral because people recognise the quicksand - the ground that held steady for years became unreliable, and thrashing harder might only pull you under faster.
Delegating Thinking to AI Erodes Mastery: But Active Use Builds It | Learning Gap 🧠
Anthropic just published research revealing a critical nuance about AI and learning. Software engineers learning a new library with AI assistance scored 17% lower on comprehension tests than those working by hand - nearly two letter grades. But AI use didn't guarantee worse outcomes. How people used it determined what they learned. Those who scored well used AI to build comprehension through follow-up questions and conceptual inquiry, not just answer generation. Those who delegated thinking entirely - what Anthropic calls "cognitive offloading" - showed the weakest mastery, particularly on questions testing deeper understanding relating to solution failure and underlying “why”.
This week's developments pose uncomfortable questions for higher education. If China can prohibit addiction-by-design and Singapore can commission national AI missions within weeks, where is the similar urgency in HE? If courts can rule that AI-generated work lacks professional protection, what does it mean to certify student capability in tasks AI performs without accountability? If research shows cognitive offloading erodes mastery while frameworks verify independent performance, are institutions measuring what matters? When thirty million people recognise the ground shifted and tech insiders describe handing entire jobs to AI, the gap between institutional response and operational reality becomes harder to ignore. The question isn't whether institutional response was rigorous enough. It's whether any institutional response can match acceleration when the capability-governance gap widens daily.
Can Students Ghost a Degree for $20/Month? | Adjunct Intelligence Season 2 🎙️
Season 2 of Adjunct Intelligence just dropped - Dale Leszczynski and I spent an hour looking back at 2025's chaos: the safety collapse, the financial ouroboros, why the plagiarism debate is dead, and whether universities are selling a product that can now be ghosted for $20/month. Next episode drops predictions for 2026. Find it wherever you listen to podcasts or watch on YouTube.