AI Research Radar

Saturday, May 23: 1 frontier paper out of 187 read.

1 frontier9 cleared187 papers read

Frontier

Papers rated as moving the field

Hallucination as Exploit

“A HACR audit on 500 stratified task keys shows that unsupported action-critical claims reach unsafe execution for naive agents (100.0%) and prompt-only defenses (49.6%), but not for ECA.”

The authors, in the abstract

Evidence-Carrying Multimodal Agents

arXiv PDF

This week in research

May 17–23: 50 cleared, 3 frontier

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

Models under test

Papers this week that name the model

GPT-5.43
GPT-4o3
Kimi K2.62
Gemini 3 Flash2
GPT-4o mini2
Claude Opus 4.62
Claude Sonnet 4.51
GPT-5.51

Labs publishing

Papers this week with an author at the lab

Amazon3
Google2
Meta1

Also cleared

8 papers

SCI-Defense: Defending Manipulation Attacks from Generative Engine OptimizationAmazon
Frequency-Domain Regularized Adversarial Alignment for Transferable Attacks against Closed-Source MLLMs
What Counts as AI Sycophancy? A Taxonomy and Expert Survey of a Fragmented Construct
On-Policy Consistency Training Improves LLM Safety with Minimal Capability Degradation
Evaluating Large Language Models as Live Strategic Agents: Provider Performance, Hybrid Decomposition, and Operational Gaps in Timed Risk Play
Benchmarking Autonomous Agents against Temporal, Spatial, and Semantic Evasions
Compiling Agentic Workflows into LLM Weights: Near-Frontier Quality at Two Orders of Magnitude Less Cost
Can AI Make Conflicts Worse? An Alignment Failure in LLM Deployment Across Conflict Contexts

SIGNIFICANT ON THE BOARDS

MODEL TRACKER2,647 new benchmark results across 174 models this week DEAL FLOW$158.1B disclosed AI funding this week across 62 rounds POLICY126 policy events tracked across 7 days