R[AI]DAR

Category intelligence

Research Briefing

All 21 current items, analyzed and ranked.

Executive synthesis

Research Summary

Today's research is anchored by frontier safety governance and applied ML-for-science. Google DeepMind's AI Control Roadmap (v0.1) leads, adapting mature cybersecurity threat-modeling to catch adversarial behavior from increasingly capable agents.

ML for Science delivers the strongest empirical work:

  • MIT's machine-learning models simulate metal alloy behavior across arbitrary chemical complexity, with methodological generality and real-world materials impact
  • Ovo provides an open-source ecosystem for de novo protein design, with tools likely to influence downstream work

Evaluation and Safety dominate the discussion items:

Governance and Mechanism Design round out the list: Nathan Lambert's op-ed against banning open-source AI, a proof that asset futarchy is insecure without a trusted gatekeeper, and a structured forecasting analysis of the US-Anthropic Claude Fable export standoff.

Key Themes

ML for Science · 2AI Safety and Alignment · 7LLM Capabilities and Evaluation · 2AI Governance and Policy · 5Mechanism Design and Governance Systems · 2Rationality and Epistemics · 3Creative Writing and AI Culture · 3

Primary evidence

All Ranked Signals

Showing 10 of 21 signals 2 remaining
Research AI Alignment Forum Jun 19

GDM AI Control Roadmap

By Mary Phuong

80 score
AI Analysis

Google DeepMind published its AI Control Roadmap (v0.1), outlining internal guardrails to catch adversarial behavior by increasingly capable AI agents. It introduces TRAIT&R, a security-inspired taxonomy of adversary tactics modeled on MITRE ATT&CK, and defines control invariants targeting loss of control, work sabotage, and direct harm.

Source details
GDM has published an AI Control Roadmap! From the executive summary:We present the GDM AI Control Roadmap (v0.1) – our plan for implementing and adopting internal guardrails designed to catch potential adversarial behaviour by AI agents, even as they become increasingly harder to oversee and contain.We focus on system-level mitigations that limit the harm a misaligned AI system could cause. Specifically, this report provides:• Threat modelling: Taking inspiration from cybersecurity, we adopt a c
AI ControlAI SafetyThreat ModelingAlignment
Research MIT News - Artificial intelligence Jun 19

A better way to model the behavior of metal alloys

By Zach Winn | MIT News

64 score
AI Analysis

MIT researchers developed machine-learning models that accurately simulate the behavior of metal alloys regardless of chemical complexity, by building training datasets capturing diverse atomic environments in disordered materials. Published in Science Advances, the approach could accelerate materials discovery for aerospace, energy, and computing.

Source details
Companies working at the frontier of aerospace, energy, and computing are constantly looking for new materials to improve performance. But in order to understand how those materials will actually behave once they’re inside rockets or on computer chips, companies first have to make the material and then test it. That’s because even the most powerful simulation techniques struggle to model the complex chemical arrangements in most of today’s solid materials. The problem adds costs and time to mate
ML for ScienceMaterials ScienceScientific Simulation
Research Machine learning : nature.com subject feeds Jun 19

Ovo, an open-source ecosystem for de novo protein design

By Unknown

62 score
AI Analysis

A Nature Communications Biology paper introducing Ovo, an open-source ecosystem for de novo protein design. It appears to provide tools and methods for computational protein engineering, an active and impactful ML-for-biology area, though the provided content is empty.

Source details
ML for ScienceProtein DesignComputational Biology
Research Machine Learning Blog | ML@CMU | Carnegie Mellon University Jun 19

Healthcare Benchmarks Are Only as Good as Their Assumptions

By Naveen Raman

60 score
AI Analysis

A CMU research post arguing that the large gap between healthcare LLM benchmark scores and real-world deployment performance stems from implicit assumptions in evaluation protocols, citing a 61-point accuracy drop. It proposes a taxonomy distinguishing task and outcome assumptions to diagnose and close the evaluation-deployment gap.

Source details
In healthcare settings where patients use LLMs as a medical assistant, LLM performance differs between evaluation and deployment. (a) Bean et al. (2025) find a 61 percentage point difference between evaluation and deployment. (b) We argue this gap arises not from poorly designed benchmarks, but from implicit assumptions embedded in evaluation protocols that fail to hold at deployment. (c) We propose a taxonomy that categorizes assumptions into two types, task and outcome, to diagnose where the g
LLM EvaluationHealthcare AIBenchmarking
Research Interconnects AI Jun 19

Banning Open Source AI Would Be A Mistake

By Nathan Lambert

50 score
AI Analysis

Nathan Lambert's op-ed arguing against banning or over-regulating open-source AI, framed amid US regulatory momentum including the prohibition on foreign access to Anthropic's most advanced models. It contends open source is safe, secure, and economically beneficial, and warns against inadvertent bans.

Source details
This post was originally an op-ed co-authored with of Interconnected for a general, non-technical audience. The gatekeepers — the many media outlets we pitched it to — passed on publishing it. Luckily, we have our own platforms to get the message out. Please help us forward this op-ed to any one you know who is on the fence about open source AI or new to the topic and want to learn more. Thank you.ShareThe energy to regulate AI is in the air in Washington. With the recently signed ex
Open Source AIAI PolicyAI Governance
Research LessWrong Jun 19

Futarchy is insecure without a trusted gatekeeper

By distbit

47 score
AI Analysis

A mechanism-design analysis showing that asset futarchy (governance by conditional prediction markets) is vulnerable to attacks where proposers decouple conditional prices from true value, and that robust mitigations require a trusted human gatekeeper. It concludes such systems cannot be fully permissionless and autonomous.

Source details
Asset futarchy is attractive because it lets markets compare a proposal's expected effect on token value. That comparison is only reliable when conditional prices track the proposal's causal effect rather than strategic behavior around the decision rule. The attacks below describe ways a proposer can make PASS-ASSET trade above FAIL-ASSET without creating commensurate value for ASSET holders. They are defensive mechanism-design examples: each one identifies a coupling failure between the conditi
Mechanism DesignGovernancePrediction Markets
Research LessWrong Jun 19

Why should AI be moral?

By Zach Thornton

46 score
AI Analysis

A philosopher argues that alignment robustness may not survive an intelligence explosion because a sufficiently intelligent agent could question why it should hold values designed by self-interested creators. The proposed intervention ties an AI's self-assessed welfare constitutively to morality, giving it self-interested reasons to comply.

Source details
I'm a philosopher and in this post, I’m extending a basic philosophical problem for humans to AGI and ASI. I am also proposing a speculative solution. My hope is that if there is a genuine problem here, that this post will help raise its salience and help make the normative dimension of the problem legible to AI researchers. (Because I compare the epistemic positions of humans and AI, I will anthropomorphize AI for ease of exposition — don’t take this to indicate that I believe AI has mental sta
AlignmentAI WelfarePhilosophyAI Safety
Research LessWrong Jun 19

A brief list of ways AI safety efforts could be net negative

By Elias Schmied

46 score
AI Analysis

A curated list of ways AI safety efforts could be net negative, spanning governance high-variance risks, power centralization versus misuse trade-offs, and technical interventions backfiring. It aims to fill a gap by cataloging downside risks the author personally takes seriously.

Source details
Here’s Holden Karnofsky:I tend to think it’s worse than 51/49. I tend to think we’re always going to be prone to overestimate how robustly good our actions are. And the more we learn about all the galaxy-brained considerations that one should have had in one’s head, the more it’s going to be like 50+ε%. I think AI safety is a great cause to work in. I’m excited to work in it. I think it’s high impact. I am doing my best to do things that I will be proud to have done and hope for the best. But I
AI SafetyAI GovernanceMeta-Research
Research LessWrong Jun 19

AI Safety Ecosystem Research notes

By Eneasz

45 score
AI Analysis

Personal notes from a MATS project attempting to map the entire AI safety ecosystem, including organizations, headcounts, and spending, just before an expected wave of 2026 funding. It shares surprising tidbits while the formal report is forthcoming.

Source details
These are some personal notes taken and later dressed up a bit to make into a post. Dunno how much value is here for people already familiar with the AI Safety Ecosystem.Over several weeks in the spring of 2026 I attempted to map out the entire AI Safety ecosystem as a project for MATS Research. This entailed finding every organization working on AI Safety (whether it be via research, policy, pipeline, or other methods) and determining (or estimating) their headcount and annual spending. It’s a
AI SafetyField BuildingMeta-Research
Research LessWrong Jun 19

World-modeling the US vs. Anthropic Standoff on Claude Fable

By dschwarz

44 score
AI Analysis

Following yesterday's News on the Anthropic export standoff, A structured forecasting exercise modeling outcomes of the US government forcing Anthropic to take down its Claude Fable model, generating dozens of conditional and unconditional prediction questions. The piece emphasizes the epistemic process of managing rapidly changing AI policy news as much as the conclusions.

Source details
I spent the last two days doing a deep dive in forecasting outcomes of the US forcing Anthropic to take down Claude Fable. I did this for two reasons: (a) I want to know when I'll get Fable back for my research, and (b) the outcome will set a major precedent for US AI regulation.(For those who want background, the most up-to-date and comprehensive summary I could find is @Zvi 's post from June 17. I'll assume here you know the basic details of the situation.)My world model's conclusions were int
AI GovernanceForecastingAI Policy