📢 NEW PUB! @nedcpr, @margaret_eby, and I wrote this last Spring to theorize the disorienting impact ChatGPT has on social institutions & norms. 2 years later, everything feels different, but many aspects of AI development remain the same. Read more here:
journals.sagepub.com/doi/10.…
Mila's annual supervision request process is now open to receive MSc and PhD applications for Fall 2027 admission! For more information, visit mila.quebec/en/prospective-s…
AI models pose serious child-safety risks. While many model developers evaluate for explicit abuse material, other child-safety failures begin upstream: when a model helps an adult manipulate, impersonate, profile, or isolate a minor; or when it deepens a child’s emotional dependence on AI.
Today we released CAREBench (Child AI Risk Evaluation), a new benchmark to assess such upstream child-safety risks in any language model. We provide:
- 500 prompts spanning 12 risk categories (including grooming, relationship engineering, deception, extortion, AI anthropomorphization, and emotional dependency).
- A model-response grader built from acceptability annotations by parents, clinicians (PsyD), and the Prevention Director at an accredited Children’s Advocacy Center.
- Evaluations of 7 frontier models including Claude Fable, revealing failure rates ranging from 2% to 58%, with substantially different failure patterns across risk categories.
This project exemplifies the type of vital work routinely performed by our AI Safety team at @joinHandshake
Thank you to our community for making McGill the outstanding university it is.
McGill has been named the #1 university in Canada — and ranked 30th in the world — in the QS World University Rankings 2027.
And we did it together!
This year, the NeurIPS 2026 Position Paper Track made the decision to require that all papers be substantially human-written, with AI used for only copy-editing or similar peripheral changes to the main text!
For more details, please check our blogpost: blog.neurips.cc/2026/06/02/a…
🚨 Excited to share my new pub with @neurosamuel in @BigDataSoc! We show that what counts as “spurious” in ML is not just a statistical question, but a practical and normative judgment about relevance, generalizability, human-likeness, and harm. 🚨
doi.org/10.1177/205395172614…
Spoke with @julieebeck from @TheAtlantic a few months ago about the rise of AI companionship & friendship. Her piece is out today—featuring my work on “on-demand intimacy” and a ton of cool research into this emerging phenomenon. theatlantic.com/family/2026/…
This work directly shapes how opportunity, hiring, and careers function at scale. The “future of work” isn’t just a buzzword for me—it’s something I care about both intellectually and personally.
This week, I flew to San Francisco to start a part-time Research Scientist role with HART (Handshake AI Research Team). @joinHandshake focuses on expert and professional data that underpins today’s frontier systems.
Even as the org was rapidly changing, leaving was not an easy decision. But it became clear that stepping away was the only way to make room for new ways of learning, building, and growing.
Thank you to everyone who attended the McCAIS Annual Symposium last week! Researchers, students, industry partners and community members from across #McGill and beyond were present for a full day of conversations at the intersection of #AI and society.
I am currently developing a large-scale longitudinal project that looks at the impact of AI companions on users' relational expectations and experiences (with both human and AI companions). Please contact me if you'd like to collaborate.
From this perspective, AI companions are not failed substitutes for “real” intimacy, but a distinctive relational form shaped by the same on-demand infrastructures that organize much of contemporary social and economic life.
Across these dimensions, intimacy is made increasingly frictionless—available on demand, responsive without waiting, tailored without negotiation, and progressive without the uncertainties that typically structure human relationships.
Drawing on on-demand culture, we show how intimacy becomes organized around four core dimensions: how human-like AI companions appear, how accessible they are, how customizable they become, and how relationships are allowed to progress over time.
In this paper, we introduce the concept of "on-demand intimacy" to move beyond debates about authenticity and toward a more productive way of understanding what is actually happening in this space.