AI Research Radar

Monday, June 1: 3 frontier papers out of 741 read.

3 frontier54 cleared741 papers read

Frontier

Papers rated as moving the field

GigaSpeechBench

“Extensive evaluations of leading foundation models and commercial APIs reveal significant performance degradation in these challenging settings, exposing critical evaluation blind spots.”

The authors, in the abstract

A Real-World Multilingual Speech-to-Text Benchmark

AlibabaBenchmarkarXiv PDF

CyberGym-E2E

Benchmark

Mellum2 Technical Report

This week in research

May 26 – Jun 1: 179 cleared, 12 frontier

The week’s papers have not been sorted into fields yet.

Models under test

Papers this week that name the model

GPT-5.45
GPT-5.25
Claude Sonnet 4.64
Gemini 3.1 Pro4
GPT-4o4
GPT-4o mini3
Claude Opus 4.63
Claude Sonnet 4.52

Labs publishing

Papers this week with an author at the lab

Alibaba4
Amazon3
Microsoft3
Google2
Tencent2
Baidu1
IBM1
NVIDIA1

Also cleared

51 papers

How Early Adopters Used Generative AI Worldwide: Variation by Country Income and LanguageMicrosoft
ConsisGuard: Aligning Safety Deliberation with Policy Enforcement in LLM GuardrailsAlibaba
NVIDIA OmniDreams: Real-Time Generative World Model for Closed-Loop Autonomous Vehicle SimulationNVIDIA
The Perils of Agency: How Developers Perceive, Prioritize, and Address Risks in Agentic AI ProductsIBM
Phantoms and Disclosures: A Statistical Framework for Auditing Privacy in Synthetic DataGoogle
BitNet Text EmbeddingsMicrosoft
Can Subgraph Explanations Be Weaponized to Steal Graph Neural Networks?Code
EMBGuard: Constructing Hazard-Aware Guardrails for Safe Planning in Embodied AgentsCode
SABER: Benchmarking Operational Safety of LLM Coding Agents in Stateful Project WorkspacesCode
Neuro-Symbolic Verification of LLM Outputs for Data-Sensitive Domains (extended preprint)
Reward Bias Substitution: Single-Axis Bias Mitigations Redirect Optimization Pressure
Inferring the Size of Large Language Models From Popular Text Memorization

All 51 papers

SANA-Streaming: Real-time Streaming Video Editing with Hybrid Diffusion Transformer
The Surface You Test Is Not the Surface That Breaks
Auditing LLM Benchmarks with Item Response Theory
CacheProbe: Auditing Prompt Cache Isolation in Gateway APIs
COFT: Counterfactual-Conformal Decoding for Fair Chain-of-Thought Reasoning in Large Language Models
BOKBO (Best of K Bad Options): Calibrated Abstention for VLA Policies
Automatically Attacking Software Reverse Engineering AI Agents
Depth-Dependent Indirect Prompt Injection in Tool-Calling ReAct Agents: Injection Depth, Payload Framing, and Turn-Budget Sensitivity
MosaicLeaks:Privacy Risks in Querying-in-the-Open for Deep Research Agents
Anchoring LLM Gender Bias to Human Baselines: A Cross-Lingual Audit
Send a SCOUT First: Pre-hoc Reasoning for Adaptive Detector Allocation in Prompt-Injection Defense
COMPASS: Cognitive MCTS-Guided Process Alignment for Safe Search Agents
From Prompt Injection to Persistent Control: Defending Agentic Harness Against Trojan Backdoors
LLM-FACETS: A Privacy-Preserving Framework for Evaluating LLM Transparency and Accountability
Divergence Decoding: Inference-Time Unlearning via Auxiliary Models
Reinforcement Learning Amplifies Emergent Misalignment from Harmless Rewards
Stateful Online Monitoring Catches Distributed Agent Attacks
Progressive in Principle, Centrist in Practice: LLM Political Bias Is Instrument-Dependent
Citation Grounding Measures the Oracle: Graph Coverage Determines Reported LLM Hallucination Rates in Law
Designing for Doubt: The Case for Informed Abstention in Autonomous Agents
Organizational Control Layer: Governance Infrastructure at the Execution Boundary of LLM Agent Systems
What If Prompt Injection Never Left? Rethinking Agent Security through Cross-Session Stored Prompt Injection
MCBench: A Multicontext Safety Assessment Benchmark for Omni Large Language Models
Where Rectified Flows Leak: Characterising Membership Signals Along the Interpolation Path
Collective Hallucination in Multi-Agent LLMs:Modeling and Defense
Poise: Position-Aware One-Instruction Skill Injection for Silent Execution on LLM Agents
When LLMs Invent Rust Crates: An Empirical Study of Hallucination Patterns and Mitigation
Online Shift Detection and Conformal Adaptation for Deployed Safety Classifiers
AI Contagion in Social Networks
AuAu: A Benchmark for Auditing Authoritarian Alignment in Large Language Models
When Agent Automation Becomes Profitable: Quantifying and Insuring Autonomous AI Risk through Trace-Economic Underwriting
Fara-1.5: Scalable Learning Environments for Computer Use Agents
Right Knowledge, Wrong Answer: Characterizing Parametric Temporal Conflict in Open-Weight Language Models
Intent-Governed Tool Authorization for AI Agents
The Watermark Shortcut: How Provenance Marking Sabotages Audio Deepfake Detection
Breaking the Rounding Trap: Securing LLMs against Quantization-Conditioned Backdoors
Nemotron-Labs-Diffusion-Image: Advancing Masked Discrete Diffusion for High-Resolution Image Synthesis
GenPage: Towards End-to-End Generative Homepage Construction at Netflix
ShopX: A Foundation Model for Intent-to-Item Fulfillment in Agentic Shopping

SIGNIFICANT ON THE BOARDS

MODEL TRACKER2,647 new benchmark results across 174 models this week DEAL FLOW$158.3B disclosed AI funding this week across 63 rounds POLICY128 policy events tracked across 7 days