iOSWorld
“Privileged vision+XML access improves frontier models by up to 26 percentage points, while smaller models do not benefit from added accessibility-tree input.”
The authors, in the abstract
A Benchmark for Personally Intelligent Phone Agents
AI Research Radar
Tuesday, June 9: 3 frontier papers out of 700 read.
Frontier
Papers rated as moving the field
How we chooseCRANE
“On the out-of-distribution MMEVOKE benchmark, CRANE reaches 87.0% under gold retrieval.”
Sycophancy Towards Researchers Drives Performative Misalignment
“First, we show that evaluation awareness persists even when we tell models they are deployed, which contradicts the scheming story which predicts less misalignment when the model perceives evaluation.”
This week in research
Jun 3–9: 145 cleared, 12 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
60 papers
- Defenses & Enablers For Skill Injection Attacks on Terminal Based AgentsNVIDIA
- Easier to Mislead Than to Correct: Harmful and Beneficial Revision in LLM ConformityAmazon
- Safety is Contextual, LLM-Judges Are Not: Navigating the Rigid Priors of EvaluatorsCohere
- CARE: A Conformal Safety Layer for Medical SummarizationGoogle
- Autonomous Incident Resolution at Hyperscale: An Agentic AI Architecture for Network OperationsMicrosoft
- Learning to Attack and Defend: Adaptive Red Teaming of Language Models via GRPOMicrosoft
- MLingualFC: Evaluating Jailbreak Vulnerabilities in Multilingual Vision-Language ModelsCode
- The Cold-Start Safety Gap in LLM AgentsCode
- Defending Against Malicious Finetuning by Scaling Train-time Adversarial AttacksCode
- Beyond FLOPs: Benchmarking Real Inference Acceleration of LLM Pruning under a GEMM-Centric TaxonomyCode
- Self-Mined Hardness for Safety Fine-Tuning
- ASH: Agents that Self-Hone via Embodied Learning
All 60 papers
- ConflictRAG: Detecting and Resolving Knowledge Conflicts in Retrieval Augmented Generation
- Post-Trained MoE Can Skip Half Experts via Self-Distillation
- Beware of GeeksBearing Gifts: Building True EU Frontier AI Sovereignty
- Syll: Open-Source Personal Automation with Cross-Surface Execution
- Can You Trust What You See? Human and AI Detection of Synthetic Legal Evidence
- Evaluation of ML Resource Utilization Requires Model Life Cycle Assessment
- Cherry-pick Override: Unsafe Directional Commitment in LLM Judges under Mixed Evidence
- Overcoming the Regulatory Bottleneck via Agent-to-Agent Protocols: A Nuclear Case Study
- The CIFAR Synthetic Evidence Corpus for Detecting AI-Generated Evidence
- Shared Latent Structures Enable Unified Backdoor Detection and Mitigation in LLMs
- RecurGuard: Runtime Monitoring for Reasoning-Token Consumption Attacks
- VATS: Exploiting Implicit Authority in Error-Path Injection via Systematic Mutation
- MC-PDD: Masked Corpus-Level Pretraining Data Detection for Black-Box Large Language Models
- EgoAERO: Learning Dexterous Manipulation from a Single Egocentric Video without Object Assets
- vla.cpp: A Unified Inference Runtime for Vision-Language-Action Models
- SciTrace: Trajectory-Aware Safety Reasoning for Scientific Discovery Agents
- RiskNet: A large-scale dataset of AI risk incidents from news with alignment and multi-dimensional annotations
- Hiding in Plain Floats: Steganographic Carriers for Indirect Prompt and Content Injection
- Sycophancy as a Multilingual Alignment Failure: How Safety Degrades Across Languages, Topics, and Models
- Testing the Black Box: Structural Barriers to Independent Evaluation of Consumer-Facing Health LLMs
- Friend or Foe? Language as an ideological switch in open-weight LLMs under Russian disinformation stress
- SpectrumKV: Per-Token Mixed-Precision KV Cache Transfer for Prefill-Decode Disaggregated LLM Serving
- Data Agents Under Attack: Vulnerabilities in LLM-Driven Analytical Systems
- Activation Steering Induces Emergent Misalignment: A More Comprehensive Evaluation
- RAILS: Verification-Native Clearing For Agentic Commerce
- The Amplifying Mirror: Locating and Steering the Partisan Direction inside a Large Language Model
- Governance Controls for AI-Generated Test Artifacts in Autonomous Software Testing
- Oversight Has a Capacity: Calibrating Agent Guards to a Subjective, Fatiguing Human
- Hardening Agent Benchmarks with Adversarial Hacker-Fixer Loops
- Document-Authored Control-Signal Impersonation: A Low-Cost Indirect Prompt Attack on RAG Safety Boundaries
- SafeRun: Enabling Determinism in LLM Planning for Running
- Context-Fractured Decomposition Attacks on Tool-Using LLM Agents: Exploiting Artifact Provenance Gaps
- Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges
- Steganography Without Modification: Hidden Communication via LLM Seeds
- Pretrained, Frozen, Still Leaking: Auditing Cross-Encoder Attribute Transfer in EEG Foundation Models
- Trustworthy Smart Fabs via Professional Proxies: Scaling Safe and Sustainable by Design (SSbD) through Industrial Data Spaces
- Brain-Prompt Injection: A Route-Safety Audit for BCI-LLM Agents
- Distilling Safe LLM Systems via Soft Prompts for On Device Settings
- Benchmarking Empirical Privacy Protection for Adaptations of Large Language Models
- Now You (Still) See Me: Detecting Evasive Steganographic Payloads in LLMs
- AI Assurance in UK Defence: Challenges in Operationalising JSP 936
- Targeting World Models to Compromise Robot Learning Pipelines
- Model Poisoning Against Federated Model Adaptation with Chain of Bit-Flips
- SecureClaw: Clawing Back Control of LLM Agents
- Safe-RULE: Safe Reinforcement UnLEarning
- Seeing the Hivemind: A Consensus-Aware Interaction Technique for Mitigating AI Homogenization
- Beyond Probabilistic Similarity: Structural, Temporal, and Causal Limitations of Retrieval-Augmented Generation in the Legal Domain
- The Neutral Mask: How RLHF Provides Shallow Alignment while Leaving Partisan Structure Intact in a Large Language Model