Thank you CBS @wbz for covering our work on teen depression on your show this evening! cbsnews.com/boston/news/head… via @wbz

Across our projects @GoogleDeepMind, we are witnessing a genuine phase transition in the pace of scientific discovery. You will see a glimpse of this in our releases in the next few weeks.

Excited to be in lovely Trier, Germany for the Minerva-Gentner Symposium on Precision in Psychotherapy. Moderating the opening panel on Precision in Measurement with a terrific group: Dana Atzil-Slonim, @aidangcw, @hadarfisher, Eva Gilboa-Schechtman & Gregor Domes.

TIPS turns 10 this December! 🎉 Join us in Boston Dec 10–11 for #TIPS2026 —two days exploring AI, digital phenotyping, wearables, VR & other emerging technologies transforming mental health research and care. ⏰ Early-bird registration ends TODAY! 🔗 lnkd.in/gCyCVAkx

Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.

A very warm welcome to Dr. Ophir Katzenelenbogen, our new postdoc in the lab! Ophir studies how mental health unfolds within close relationships, focusing on dyadic/interpersonal processes that shape risk and resilience, with the goal of informing more personalized interventions.

Introducing Griffin, the first model to pass the video Turing test. 48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video. It’s the first Human Interaction Model (HIM).

I guess the good news is that has only two dimensions. For now...

So excited to return to SRP, for my first time as a non-trainee 🥺 Come keep me company during my talk on symptom-specific speed of improvement during psychotherapy (work with @FabienneMink @ChrisWebbPhD and Lutz) @srp_science #SRP2026

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training. You can read the full post here: darioamodei.com/post/we-must…

3-day intensive #fMRI course *live online* Sept 9-11, 2026. #SPM, #ICA, GLM, connectivity, mediation, MRI physics, #DataScience with @vcalhoun.bsky.social and Kent Kiehl. We love talking methods & connecting with colleagues! Come join us! Register here: torwager.github.io/mindfmric…

Honored that the @BBRFoundation featured our work in its Fall 2026 issue. The piece highlights a couple of examples of how we’re combining smartphones, wearable sensing, and AI to better understand adolescent depression and personalize treatment. tinyurl.com/24h47pfa

METR & Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.

This is probably the most well done RCT on meditation I’ve ever seen

In a new preregistered RCT w/ US & UK samples (N > 5k) led by Otto Simonsson, three brief meditation practices modestly beat a control on negative affect. A multi-arm causal forest then identified who benefits from which practice (d=0.16, up to d=0.43). doi.org/10.1016/j.brat.2026.…

Many thanks to the @Google Cambridge team for hosting me and inviting me to speak about our work on AI and digital mental health in youth. Great discussions, inspiring work, beautiful spaces, a putting green, great snacks, and even an automated mouthwash dispenser ;-) Thank you!

Yale Psychology is hiring an Assistant Professor in Clinical Psychology! Review of applications will begin on September 2, 2026. Link: apply.interfolio.com/188806 Please share widely & apply!

Photo from our recent lab dinner. Science is a team sport, and I feel so fortunate to work with such an amazing group of colleagues and trainees. In the end it’s the people who make this work meaningful. Grateful for this team and excited for what’s ahead. Cheers! 🍻

Looking forward to SPR 2026 in Osaka 🇯🇵 next week, where members of our lab will be participating in several sessions focused on AI-informed interventions, digital mental health, personalization, measurement, and psychotherapy process. If you’ll be at SPR swing by & say hello!

Using causal ML and data from ~90k NHS patients, we found that only 31% received the treatment the algorithm identified as optimal. Recovery rates were 11% pts higher among patients with the strongest recommendations who received their indicated treatment lnkd.in/e5Muc4BJ

I dug into NSDUH 2004-2024 data on depression in U.S. teens (12-17). Rates increased far more sharply for girls than boys, peaking around COVID before modest declines. Shaded region = % with depression receiving mental health treatment. Both depression and unmet need remain high.