SEC-bench Pro
“Our evaluation shows that coding agents with frontier models remain below 40% success on both evaluated engines.”
The authors, in the abstract
Can Language Models Solve Long-Horizon Software Security Tasks?
AI Research Radar
Wednesday, May 27: 2 frontier papers out of 260 read.
Frontier
Papers rated as moving the field
How we chooseAlignment Tampering
“Our experiments demonstrate amplification across diverse biases: from keyword bias to propaganda (e.g., sexism), brand promotion, and instrumental goal-seeking.”
This week in research
May 21–27: 99 cleared, 7 frontier
The week’s papers have not been sorted into fields yet.
Models under test
Papers this week that name the model
Labs publishing
Papers this week with an author at the lab
Also cleared
18 papers
- Large Language Model-Powered Query-Driven Event Timeline Summarization in Industrial SearchBaidu
- Natural Language Query to Configuration for Retrieval AgentsMicrosoft
- GraphIP-Bench: How Hard Is It to Steal a Graph Neural Network, and Can We Stop It?Code
- Eroding Trust in Real Speech: A Large-Scale Study of Human Audio Deepfake Perception
- MemMorph: Tool Hijacking in LLM Agents via Memory Poisoning
- TSFMAudit: Data Contamination Auditing in Forecasting Time Series Foundation Models
- A Universal Cliff and a Design Fingerprint: Cross-Section Defect Detection Under LLM Orchestration
- Erased but Exploitable: Black-box Embedding-Aware Prompting Against Unlearned Text-to-Image Diffusion Models
- Memory Architectures for Multi-Turn Text-to-SQL: A Benchmark and Empirical Study
- Vectors Are Not Neutral: Sensitive-Information Inference from Exported LLM Representations in Summarization
- LURE: Live-Usage Replay Evaluations for Reducing Evaluation Awareness
- Open-Weight LLM Fine-Tuning Defenses are Susceptible to Simple Attacks
All 18 papers
- AI evaluation may bias perceptions: The importance of context in interpreting academic writing
- EmoDistill: Offline Emotion Skill Distillation for Language Model Agents in Adversarial Negotiation
- Black-box Membership Inference Attacks on the Pre-training Data of Image-generation Models
- The Coverage Illusion: From Pre-retrieval Routing Failure to Post-retrieval Cascades in a Production RAG System
- FinHarness: An Inline Lifecycle Safety Harness for Finance LLM Agents
- GENESIS: Harnessing AI Agents for Autonomous 6G RAN Synthesis, Research, and Testing