AI agents can now manipulate your organization. Are you ready?
SPONSORED POST: Agents with hands require a hands-on policy
America Forever Bytes
agents · AFBytes
Other
agents
Recent coverage
Only 11% of production agents pass the AI agent security bar
AI agent security lags capability. A new report scores 100 production agents and finds 98% carry the lethal trifecta of attack conditions.
[2606.03565] Skill Is Not Document: A Query-Conditional Benchmark and Two-Stage Retriever for LLM Agent Skill Routing
Abstract page for arXiv paper 2606.03565: Skill Is Not Document: A Query-Conditional Benchmark and Two-Stage Retriever for LLM Agent Skill Routing
[2606.02871] Adaptive Latent Agentic Reasoning
Abstract page for arXiv paper 2606.02871: Adaptive Latent Agentic Reasoning
[2606.02875] Handoff Debt: The Rediscovery Cost When Coding Agents Take Over Interrupted Tasks
Abstract page for arXiv paper 2606.02875: Handoff Debt: The Rediscovery Cost When Coding Agents Take Over Interrupted Tasks
[2606.03034] Capability Advertisement as a Market for Lemons: A Trust Layer for Heterogeneous Agent Networks
Abstract page for arXiv paper 2606.03034: Capability Advertisement as a Market for Lemons: A Trust Layer for Heterogeneous Agent Networks
[2606.03374] eMEM: A Hybrid Spatio-Temporal Memory System For Embodied Agents
Abstract page for arXiv paper 2606.03374: eMEM: A Hybrid Spatio-Temporal Memory System For Embodied Agents
[2606.03544] SAGE: A Quantitative Evaluation of Socialized Evolution in Agent Ecosystems
Abstract page for arXiv paper 2606.03544: SAGE: A Quantitative Evaluation of Socialized Evolution in Agent Ecosystems
Do AI Agents Outnumber Humans On Earth? When Will They Form Societies?
At what point will the number of artificially intelligent “agents” exceed the number of humans on earth? What does that mean for how they will interact?
[2606.01886] Absorbing Complexity: An Interaction-Native Knowledge Harness for Financial LLM Agents
Abstract page for arXiv paper 2606.01886: Absorbing Complexity: An Interaction-Native Knowledge Harness for Financial LLM Agents
[2606.02304] Unified Context Evolution for LLM Agents
Abstract page for arXiv paper 2606.02304: Unified Context Evolution for LLM Agents
[2606.02314] Discovering Agents for Discovery: The Case for DNS
Abstract page for arXiv paper 2606.02314: Discovering Agents for Discovery: The Case for DNS
[2606.02388] Policy and World Modeling Co-Training for Language Agents
Abstract page for arXiv paper 2606.02388: Policy and World Modeling Co-Training for Language Agents
![[2606.02404] K-BrowseComp: A Web Browsing Agent Benchmark Grounded in Korean Contexts](/static/img/curated/desks/afbytes-asia.png)
[2606.02404] K-BrowseComp: A Web Browsing Agent Benchmark Grounded in Korean Contexts
Abstract page for arXiv paper 2606.02404: K-BrowseComp: A Web Browsing Agent Benchmark Grounded in Korean Contexts
[2606.02449] HLL: Can Agents Cross Humanity's Last Line of Verification?
Abstract page for arXiv paper 2606.02449: HLL: Can Agents Cross Humanity's Last Line of Verification?
[2606.01528] Joint Agent Memory and Exploration Learning via Novelty Signals
Abstract page for arXiv paper 2606.01528: Joint Agent Memory and Exploration Learning via Novelty Signals
[2606.01533] Multi-Agent Computer Use
Abstract page for arXiv paper 2606.01533: Multi-Agent Computer Use
One prompt, four (sub)agents, and ninety seconds to get a working app
I'm been thinking a lot about agent teams. You know, a set of AI agents that work together towards a goal. You might implement this within a single agent harnes...
Everyone Is Sharing the AI Toolkit. No One Is Asking What’s Underneath It.
A frontline agent is not an agent-based model. The difference is why your AI will fail.
[2605.31042] From Prompt Injection to Persistent Control: Defending Agentic Harness Against Trojan Backdoors
Abstract page for arXiv paper 2605.31042: From Prompt Injection to Persistent Control: Defending Agentic Harness Against Trojan Backdoors
Challenging AI Assumptions
Let’s think about centralized intelligence assumptions, advocating collaborative, decentralized, biologically inspired agent ecosystems instead.
AI Agent Infrastructure: Why Enterprises Need a Control Plane
There needs to be a solution for a control plane before corporations can scale autonomous agents from demos into live production.
Why Autonomous AI Systems Require Continuous Verification
Once an agent can execute tool calls, they require continuous oversight and runtime verification.
[2605.29744] Why Specialist Models Still Matter: A Heterogeneous Multi-Agent Paradigm for Medical Artificial Intelligence
Abstract page for arXiv paper 2605.29744: Why Specialist Models Still Matter: A Heterogeneous Multi-Agent Paradigm for Medical Artificial Intelligence
[2605.29893] Redundant or Necessary? A Benchmark for Detecting Redundant Steps in Agent Trajectories
Abstract page for arXiv paper 2605.29893: Redundant or Necessary? A Benchmark for Detecting Redundant Steps in Agent Trajectories
[2605.29440] SkillBrew: Multi-Objective Curation of Skill Banks for LLM Agents
Abstract page for arXiv paper 2605.29440: SkillBrew: Multi-Objective Curation of Skill Banks for LLM Agents
[2605.29463] Honest Lying: Understanding Memory Confabulation in Reflexive Agents
Abstract page for arXiv paper 2605.29463: Honest Lying: Understanding Memory Confabulation in Reflexive Agents
[2605.30152] Do Proactive Agents Really Need an LLM to Decide When to Wake and What to Anchor?
Abstract page for arXiv paper 2605.30152: Do Proactive Agents Really Need an LLM to Decide When to Wake and What to Anchor?
[2605.29625] Improving Collaborative Storytelling with a Multi-Agent Framework Based on Large Language Models
Abstract page for arXiv paper 2605.29625: Improving Collaborative Storytelling with a Multi-Agent Framework Based on Large Language Models
[2605.27419] APS: Bias-Controlled Adaptive Prototype Simulation for Population-Scale LLM Agents
Abstract page for arXiv paper 2605.27419: APS: Bias-Controlled Adaptive Prototype Simulation for Population-Scale LLM Agents