Project APE: Can policy evaluation be automated? Or is hallucinated slop unavoidable? Let's find out.

📢 New paper: Forecasts of explosive AI progress hinge on AI agents automating AI research. But most evaluations of agents conducting AI research focus on narrow, verifiable tasks. Can AI agents… | Sayash Kapoor

📢 New paper: Forecasts of explosive AI progress hinge on AI agents automating AI research. But most evaluations of agents conducting AI research focus on narrow, verifiable tasks. Can AI agents conduct open-ended research? https://lnkd.in/gfP-q4CD We gave agents research questions from two unpublished papers, six days, and thousands of dollars of API credits and compute. The authors of the original papers reviewed the AI-generated papers. They unambiguously rejected agents' outputs. Agents were fluent at most *engineering* tasks. They conducted serious literature reviews, debugged GPU environments, ran hundreds of experiments, and turned in camera-ready LaTeX without human help. We also found no evidence of reward hacking. If anything, we found the opposite: the agents started with marketable claims and walked them back to negative results as the evidence came in. But neither agent output was close to the bar of a top conference paper. Both papers suffered from similar failures: poor judgment about the bar for an AI paper submitted to a top conference, the lack of creative problem solving and ineffective backtracking, poor awareness of resources, and instruction drift. This research design has many limitations: the small sample size, non-blind reviews, and the reviewers knowing that the work was AI-generated. We also couldn't test Anthropic's strongest model, because Fable 5 is deliberately limited on frontier AI research tasks, so ended up using OpenClaw with Opus 4.8 (extra-high) for our main experiments and Codex with Sol 5.6 (ultra) for a robustness check. But we think the research design is still helpful in assessing AI agents' ability to conduct research, and it is complementary to evaluations on verifiable tasks, as well as blinded reviews of AI outputs. In follow-up studies, we are expanding the set of non-public papers we evaluate. If you are an AI researcher with unpublished papers, we would love to collaborate with you on our next evaluation. Expression of interest: https://lnkd.in/gpeykJea We also release the agent logs and all the code and data, so that others can conduct their own analyses of our results: https://lnkd.in/gJarPAnb Finally, we plan to conduct such evaluations regularly, and are hiring a senior researcher to help lead these efforts. Apply here: https://lnkd.in/erJZdmve I'm grateful for the core team leading this effort: Peter Kirgis, Andrew Schwartz, Stephan Rabanser, and Arvind Narayanan, and to our collaborators who reviewed AI papers, analyzed agents logs, and gave feedback on the paper: David Demitri Africa, Konstantinos V., Viet Nguyen, Dr Toby D. Pilditch, Magda Dubois, Harry Coppock, Cozmin Ududec, Nitya Nadgir, Matilda Orona, Tilman Bayer, Derrick Chan-Sew, Eric (Yue) Ling, Abhishek Shetty, Helen Toner, Gillian K. Hadfield, Seth Lazar, Steve Newman, Shoshannah Tekofsky, Rishi Bommasani

No. 002: Hallucinated citations at scale, ICML catches AI reviewers, scholarly infrastructure gaps - The Hallway Track

The Hallway Track No. 002 - How AI is affecting the practice of science

Can AI agents conduct open-ended AI research? Early evidence from two case studies

Forecasts of explosive AI progress hinge on AI agents automating AI research. But evidence on whether agents can carry out open-ended AI research is thin. Current evaluations either test agents on narrow, verifiable tasks, which excludes open-ended research, or submit AI-generated papers to blind peer review, which is overstretched, stochastic, and suffers from poor review quality. We introduce a third way to measure progress towards AI R\&D automation. An agent takes on the central, open-ended research question of a high-quality unpublished paper, and the paper's original authors grade its output. We call these shadow evaluations. We ran shadow evaluations on two unpublished NeurIPS 2026 submissions, giving frontier agents six days and thousands of dollars of compute. The agents completed all of the engineering without human help, yet could not make substantial progress towards answering the research questions. As a result, both papers were unambiguously rejected by the authors. We identify five recurring failure modes: poor judgment about the bar for publishable research, uncreative responses to shortcomings in the research design, ineffective backtracking from dead ends, poor resource awareness, and instruction drift. A robustness check with a second model and scaffold reproduced these failures. We release the expert reviews, survey responses, agent repositories, and logs. Our results provide early evidence that today's agents can do the engineering of AI research, but struggle with critical parts of the research lifecycle.

Verification in the Age of AI Scientists

AI Scientists now operate at a scale that outpaces human capacity for manual review. Systems such as Sakana’s AI Scientist write entire workshop papers end-to-end. Lila Sciences runs autonomous ``AI Science Factories’’ that hypothesize, experiment, and iterate without human guidance. FutureHouse’s Kosmos and Robin generate thousands of candidate hypotheses in a single run, and Google’s Co-Scientist proposes testable experiments at a rate no laboratory can fully evaluate. Each of these systems emphasizes verified results, yet that standard ranges from near-perfect formal proof in mathematics to decade-long clinical trials in medicine, with no shared framework for judging sufficiency across domains. As outputs scale beyond what humans can manually inspect, the question of which results to trust becomes as hard as generating them. The bottleneck for AI for Science is no longer hypothesis generation, it is verification.

Mechanism Experiments and Policy Evaluations

Randomized controlled trials are increasingly used to evaluate policies. How can we make these experiments as useful as possible for policy purposes? We argue greater use should be made of experiments that identify the behavioral mechanisms that are central to clearly specified policy questions, what we call "mechanism experiments." These types of experiments can be of great policy value even if the intervention that is tested (or its setting) does not correspond exactly to any realistic policy option.

Sharing the Algorithm: The Tax Solution to Generative AI

This article argues that tax policy offers a core tool for mitigating the sweeping public policy challenges of generative Artificial Intelligence ("AI"

How to build an AI scientist: first peer-reviewed paper spills the secrets

AI Scientist, an autonomous research tool, first released in 2024, has now undergone peer review, highlighting its strengths and limitations.

AI agents can't yet do open-ended AI research

Early evidence from two case studies

Cosmik Updates: February 2026 - Cosmik Labs

@atproto.science @cosmik.network Raising a question for the ATProto science community: Can AI agents be legitimate participants in research ecosystems? What would make their outputs trustworthy?

No. 014: Science's AI productivity paradox, Sakana clears peer review, AI proof search on open Erdős problems - The Hallway Track

The Hallway Track No. 014 - How AI is affecting the practice of science

The AI "Evaluation Crisis" Is an Opportunity to Get Data Flow Right

Why the AI evaluation crisis could force a reckoning on dataset provenance, attribution, and consent.

Society-in-the-loop: programming the algorithmic social contract

Recent rapid advances in Artificial Intelligence (AI) and Machine Learning have raised many questions about the regulatory and governance mechanisms for autonomous machines. Many commentators, scholars, and policy-makers now call for ensuring that algorithms governing our lives are transparent, fair, and accountable. Here, I propose a conceptual framework for the regulation of AI and algorithmic systems. I argue that we need tools to program, debug and maintain an algorithmic social contract, a pact between various human stakeholders, mediated by machines. To achieve this, we can adapt the concept of human-in-the-loop (HITL) from the fields of modeling and simulation, and interactive machine learning. In particular, I propose an agenda I call society-in-the-loop (SITL), which combines the HITL control paradigm with mechanisms for negotiating the values of various stakeholders affected by AI systems, and monitoring compliance with the agreement. In short, ‘SITL = HITL + Social Contract.’

AI peer review goes pro

A what we're reading spotlight