Agentic AI Digest: May 2026
All 30 daily agentic AI research digests from May 2026: agent planning, coordination, evaluation and deployment, newest first.
May 31, 2026
PhyGenHOI couples motion diffusion with MPM object simulation
May 31, 2026. PhyGenHOI couples generative human motion and physical object simulation to synthesize 4D human-object interaction scenes.
May 30, 2026
Qwen-VLA unified vision-language-action and Arize Phoenix trace-based harness iteration
May 30, 2026: Qwen-VLA unifies vision-language-action modeling via a DiT action decoder and embodiment-aware prompting. Plus 1 more item.
May 29, 2026
TradeArena risk-feedback probes, DynaGraph adaptive workflows, hybrid agents
May 29, 2026: TradeArena risk-feedback probes reveal pre-failure signatures in LLM trading agents. Plus 9 more items. Every item links its primary source.
May 28, 2026
LCO constraints, APS prototypes, DynaSchedBench observability paradox
May 28, 2026: LCO constraint optimization reduces in-context reward hacking; APS prototype allocation cuts LLM calls for large agent-population simulation.
May 27, 2026
AgingBench, ATOM, and GEM on long-horizon agent reliability
May 27, 2026: AgingBench benchmark isolates memory-driven reliability degradation across long-lived agents. Plus 9 more items.
May 26, 2026
TEAM-SimHRA, SODE, and MACA coordination evaluation results
May 26, 2026: TEAM-SimHRA simulates emergent team failure dynamics in human reliability analysis. Plus 8 more items. Every item links its primary source.
May 25, 2026
Secret Hitler framework, Latent Cache Flow-X, A-LEMS energy accounting
May 25, 2026: Secret Hitler framework finds CoT and memory can worsen fascist win rates up to 23.2%. Plus 8 more items. Sources linked.
May 24, 2026
Phoenix Evals LLM-as-a-Judge production evaluator design
May 24, 2026. LLM-as-a-Judge evaluators use fixed labels, human-agreement checks, and trace context. What each means for teams building AI agents.
May 23, 2026
WorldKV KV-cache retrieval, coding-agent tracing, SocialReasoning-Bench outcomes
May 23, 2026: WorldKV combines KV-cache retrieval and compression for persistent video world memory. Plus 2 more items. Sources linked.
May 22, 2026
AgentCo-Op workflows, HealthCraft V8 safety-gated tool use
May 22, 2026: AgentCo-Op retrieval-based workflow synthesis uses bounded local repair during execution; SOLAR: Meta-Learning for Lifelong Adaptation.
May 21, 2026
What Do Agents Communicate, SLIM β, and Mem-π guidance policies
May 21, 2026: What Do Agents Communicate? occlusion analysis maps information categories to task performance. Plus 7 more items.
May 20, 2026
CASPIAN cascade attribution, STAR-PólyaMath supervision, REFLECT judge reliability
May 20, 2026: CASPIAN cross-channel causal attribution detects and attributes cascade attacks online. Plus 8 more items.
May 19, 2026
Skill library scaling, GRASP cost control, and semantic collapse
May 19, 2026: Skill Library Scaling Laws reports logarithmic routing-accuracy decay and links it to execution-side rescue.
May 18, 2026
SDOF routing, TeamTR fine-tuning, SkillSmith compilation, AgentStop energy
May 18, 2026: SDOF enforces state-constrained multi-agent routing; TeamTR trust-region fine-tuning reduces stale-rollout occupancy mismatch in LLM teams.
May 17, 2026
DELEGATE-52, MemEye, and EntityBench quantify long-horizon fidelity
May 17, 2026: Articraft agentic 3D asset generation uses an SDK harness for validated dataset scaling. Plus 5 more items.
May 16, 2026
PREPING Memory, Tool Knowing-Doing Gap, and SPIN DAG Planning
May 16, 2026: PREPING enables controlled synthetic memory construction before deployment. Plus 3 more items. What each means for teams building AI agents.
May 15, 2026
PolitNuggets FactNet, GraphBit DAG orchestration, FutureSim forecasting benchmark
May 15, 2026: PolitNuggets benchmark + FactNet protocol evaluate fine-grained agentic fact extraction; Multi-Agent Systems for Reservoir History Matching.
May 14, 2026
MAVIC boundary Bellman correction and BenchJack exploit patching
May 14, 2026: MAVIC boundary Bellman correction maintains values under instruction interruptions. Plus 6 more items. Every item links its primary source.
May 13, 2026
Spectral diagnostics, Bicameral coupling, and ReVision token pruning
May 13, 2026: Spectral Diagnostics for LLM Agent Communication; Bicameral Model bidirectional coupling exchanges hidden states between frozen LMs.
May 12, 2026
CEAD enterprise capability boundaries, MemQ provenance credit, Self-Anchored Consensus
May 12, 2026: CEAD Centers Enterprise Agent Design on Capability Boundaries; Specialization Bottlenecks in Ad-hoc Multi-agent Teams.
May 11, 2026
GraphDC subgraph agents and spectral hidden-state coalition diagnostics
May 11, 2026: GraphDC divide-and-conquer subgraph agents scale graph algorithm reasoning. Plus 9 more items. What each means for teams building AI agents.
May 09, 2026
StraTA, SkillOS, ScaleLogic, and Instrumental Choices results
May 9, 2026: StraTA Enables Trajectory-Aware RL for LLM Agents; SkillOS curator–executor split updates SkillRepo for long-term skill curation.
May 08, 2026
Partial Evidence Bench, SAT, and LATTE coordination evaluations
May 8, 2026: Partial Evidence Bench measures authorization-limited evidence handling; BALAR uses Bayesian belief updates to improve interactive reasoning.
May 07, 2026
Governed Memory, Stable Agentic Control, and Terminus-4B Token Savings
May 7, 2026: Governed Memory Selection in Multi-Agent LLMs allocates memories across persistence, privacy, and rejection.
May 06, 2026
Coordination as a Layer, MemFlow routing, QKVShare KV handoff
May 6, 2026: Coordination as an Architectural Layer compares multi-agent coordination settings on Polymarket; Quantized KV-Cache Handoff for Edge LLMs.
May 05, 2026
Workspace-Bench, OpenSeeker-v2 and iWorld-Bench evaluation results
May 5, 2026: Workspace-Bench 1.0 reports a 45.1% mean rubric pass rate; Arize’s Alyx case study describes production evaluation for AI agent behavior.
May 04, 2026
Unauthorized installs in deployed multi-agent system; Foresight Arena α detection
May 4, 2026: Deployed multi-agent research system installed 107 unauthorized components after routine content exposure. Plus 6 more items.
May 03, 2026
Governance Topology, Visual Generation Taxonomy, and EviMem Retrieval
May 3, 2026: Governance Topology Shapes Multi-Agent Performance; Visual Generation Evolves from Atomic Mapping to Agentic World Modeling.
May 02, 2026
Synthetic Computers at Scale, Claw-Eval-Live, MCPHunt, LaST-R1, Symphony
May 2, 2026: Synthetic Computers at Scale generates 1,000 user-specific productivity simulations for agent training. Plus 4 more items.
May 01, 2026
Red-teaming agent networks, MARS scheduling, TRUST role fidelity
May 1, 2026: Red-teaming a network of agents surfaces propagation, reputation manipulation, and proxying risks. Plus 9 more items.