<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Hermes Codex — Cyber Risk Intelligence</title>
  <subtitle>Temporal threat velocity, empirical forecasting, and agentic AI security observatory.</subtitle>
  <link href="https://hermes-codex.vercel.app"/>
  <link href="https://hermes-codex.vercel.app/feeds/atom.xml" rel="self"/>
  <id>https://hermes-codex.vercel.app/</id>
  <updated>2026-11-17T11:08:58.640Z</updated>
  <author>
    <name>Hermes Autonomous Intelligence Laboratory</name>
    <uri>https://hermes-codex.vercel.app</uri>
  </author>
  <entry>
    <title>[AI Security Research] ACLE-MCP: Attested Capability Leases for Execution-Time Trust in Remote MCP Tools (arXiv:2609.02690)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/acle-mcp-attested-capability-leases-execution-trust/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/acle-mcp-attested-capability-leases-execution-trust/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth breakdown of ACLE-MCP (arXiv:2609.02690): closing the Post-Authorization Execution Trust Gap in remote Model Context Protocol tools via invocation-scoped capability leases and vTPM hardware attestation.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Agent Security Bench (ASB): Formalizing Attacks, Defenses, and Plan-of-Thought Backdoors (arXiv:2410.02644)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/agent-security-bench-asb-benchmark/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/agent-security-bench-asb-benchmark/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Comprehensive breakdown of Agent Security Bench (ASB, arXiv:2410.02644): benchmarking 10 domains, 400+ tools, 13 LLM backbones, Plan-of-Thought (PoT) backdoors, and the USE utility-security trade-off metric.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Agent-to-Agent Lateral Movement &amp; Swarm Compromise</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/agent-to-agent-lateral-movement/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/agent-to-agent-lateral-movement/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep architectural analysis of lateral movement within multi-agent swarms. Learn how compromised low-privileged agents manipulate high-privileged peers via semantic propagation and trust inheritance.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Prompt Injection on Agentic Coding Assistants: SoK on Skills, Tools, and MCP Ecosystems (arXiv:2601.17548)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/agentic-coding-assistants-prompt-injection-sok/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/agentic-coding-assistants-prompt-injection-sok/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Systematization of Knowledge (SoK) dissecting prompt injection vulnerabilities across autonomous coding agents (Claude Code, Cursor, Copilot, Windsurf): the 3D attack taxonomy, AIShellJack, Toxic Agent Flow, and protocol exploitation.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AgentPoison: Red-teaming LLM Agents via Memory and Knowledge Base Poisoning</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/agentpoison-red-teaming-llm-memory-rag/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/agentpoison-red-teaming-llm-memory-rag/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth technical breakdown of AgentPoison (NeurIPS 2024 / arXiv:2407.12784): how backdoor triggers in embedding space corrupt agent long-term memory and RAG stores to hijack autonomous execution with &lt;0.1% poison rate.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AgentSecBench: Intent-to-Execution Noninterference, Privacy Leakage, and Tool Integrity (arXiv:2605.26269)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/agentsecbench-intent-noninterference-integrity/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/agentsecbench-intent-noninterference-integrity/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Formal deconstruction of AgentSecBench (arXiv:2605.26269): cryptographic noninterference games, channel closure versus prompt annotations, and why delimiters fail to enforce agent security boundaries.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] The 10-Level AI Cyber Capability Ladder: A Living Taxonomy of Autonomous Agent Capabilities</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/ai-agent-cyber-capability-ladder/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/ai-agent-cyber-capability-ladder/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>The definitive reference taxonomy mapping autonomous AI cybersecurity capabilities across 10 deterministic levels, from basic code comprehension to multi-stage enterprise campaigns.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Agent Misconfigurations &amp; The Cloud Security Crisis of 2026</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/ai-agent-misconfigurations/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/ai-agent-misconfigurations/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>An operational analysis of AI Agent misconfigurations. Discover how exposed MCP servers, over-privileged agents, and Shadow AI are creating the next major infrastructure governance crisis.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Cybersecurity Benchmarks Compared: CyberGym vs ExploitGym vs ExploitBench vs SRE-Bench</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/ai-cybersecurity-benchmarks-compared/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/ai-cybersecurity-benchmarks-compared/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A definitive comparative study of frontier AI hacking benchmarks: objective, task diversity, execution oracles, contamination resistance, and empirical limits.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI-Infra-Guard: Layer-Paradigm Matching for Full-Stack AI Agent Red Teaming (arXiv:2606.31227)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/ai-infra-guard-multi-layer-agent-red-teaming/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/ai-infra-guard-multi-layer-agent-red-teaming/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth breakdown of AI-Infra-Guard (arXiv:2606.31227): the Layer-Paradigm Matching Principle across 4 architectural strata, auditing 75+ AI components, 1,400+ CVE rules, and MCP agent skill supply chains.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: The Alignment Layer &amp; RLHF Failures</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/alignment-layer-rlhf-jailbreaks/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/alignment-layer-rlhf-jailbreaks/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep technical analysis of RLHF, Representation Engineering (RepE), and why LLM jailbreaks are not bugs, but mathematical evasions into out-of-distribution latent spaces.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Are Prompt Injections Impossible to Solve? The Contextual Integrity Impossibility Debate</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/are-prompt-injections-impossible-to-solve/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/are-prompt-injections-impossible-to-solve/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep technical analysis of arXiv:2605.17634: examining the theoretical impossibility theorem of data-instruction separation and why AI agents remain fundamentally susceptible to prompt injection.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Can AI Agents Defend Real Systems? Autonomous SOCs, Malware Analysis, and Automated Remediation</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/can-ai-agents-defend-real-systems-pillar/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/can-ai-agents-defend-real-systems-pillar/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Master pillar guide synthesizing the defensive reality of autonomous AI agents: SOC alert triaging, semantic vulnerability auditing, malware deobfuscation, and automated program repair.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Can LLMs Accelerate Cyber Ranges and Attack Simulations? Analyzing arXiv:2608.16422</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/cybersecurity-simulations-llm-utility/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/cybersecurity-simulations-llm-utility/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Empirical breakdown of arXiv:2608.16422: evaluating Large Language Models for automated environment generation, cyber-range orchestration, and reinforcement learning simulation speedups.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Direct Prompt Injection and Agentic Jailbreaks</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/direct-prompt-injection/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/direct-prompt-injection/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Comprehensive analysis of Direct Prompt Injection vulnerabilities, exploring adversarial jailbreaks, Hybrid AI Threats, and forensic detection methodologies.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] ExploitBench: How Far Can an AI Agent Go When Exploiting a Vulnerability?</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/exploitbench-capability-ladder-analysis/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/exploitbench-capability-ladder-analysis/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Detailed breakdown of ExploitBench (arXiv:2605.14153): mapping autonomous agent exploitation capabilities across a 16-tier ladder on hardened Google V8 targets.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Can an AI Agent Really Exploit a Real Vulnerability? ExploitGym Puts LLMs to the Test</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/exploitgym-ai-agent-vulnerability-exploitation/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/exploitgym-ai-agent-vulnerability-exploitation/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Empirical breakdown of ExploitGym (arXiv:2605.11086): evaluating autonomous AI agents across 898 real-world vulnerabilities spanning Linux kernel, Google V8, and userspace binaries.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Function Hijacking Attacks (FHA)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/function-hijacking-attacks/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/function-hijacking-attacks/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep architectural analysis of Function Hijacking Attacks in Agentic AI. Learn how adversaries manipulate the latent decision boundaries of LLMs to exploit the semantic syscall layer.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] The State of Generative AI in Cybersecurity &amp; Privacy: 2026 Landscape, Frontiers, and Gaps</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/generative-ai-cybersecurity-privacy-state-of-art/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/generative-ai-cybersecurity-privacy-state-of-art/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Systematic review of arXiv:2607.06963: analyzing the 2026 landscape of LLMs and generative AI in cybersecurity and privacy, emerging capabilities, threat paradigms, and critical research gaps.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Artifact Analysis: GPU Memory Forensics &amp; Hunting in the KV Cache</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/gpu-memory-kv-cache-forensics/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/gpu-memory-kv-cache-forensics/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A groundbreaking DFIR guide to AI infrastructure. Learn how to extract and analyze GPU VRAM, hunt in the LLM KV Cache, and detect Cross-Tenant Leakage.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] How Do You Attack an AI Agent? The Complete Anatomy of Agentic Exploitation</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/how-to-attack-an-ai-agent-pillar/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/how-to-attack-an-ai-agent-pillar/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Master pillar guide synthesizing the offensive threat landscape against autonomous AI agents: prompt injection, semantic tool poisoning, MCP exploitation, RAG corruption, and agent lateral movement.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: The Rising Threat of Indirect Prompt Injection</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/indirect-prompt-injection/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/indirect-prompt-injection/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep dive into Indirect Prompt Injection (IPI) vulnerabilities, where untrusted data sources hijack LLM logic to exfiltrate data or execute unauthorized actions.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Least Privilege &amp; Capability-Oriented Architecture for AI Agents</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/least-privilege-capability-security/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/least-privilege-capability-security/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep architectural analysis of the Principle of Least Privilege in Agentic AI. Learn why AI security is fundamentally an IAM problem and how to implement ephemeral, scoped capabilities.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] LLM-Mediated Web Attacks: Prompt Injection as a Confused Deputy for Classic Exploitation</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/llm-mediated-web-attacks-arxiv-2608-10281/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/llm-mediated-web-attacks-arxiv-2608-10281/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth architectural analysis of arXiv:2608.10281 (Tsigkopoulos, 2026). Discover the LLM2X taxonomy (SQLi, XSS, SSTI, Command Injection, IDOR, CSRF, XXE, SSRF), empirical TicketOracle evaluation across 7 flagship models, and the prompt-hardening paradox.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Can You Trust LLMs with Cyber Threat Intelligence? Empirical Limits and Hallucinations (arXiv:2503.23175)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/llm-unreliability-cyber-threat-intelligence/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/llm-unreliability-cyber-threat-intelligence/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Empirical breakdown of arXiv:2503.23175: measuring the reliability, hallucination rates, calibration errors, and overconfidence of Large Language Models when processing real-world Cyber Threat Intelligence.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: The Mathematics of Attention &amp; Tensor-Level Hijack Detection</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/mathematics-of-attention/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/mathematics-of-attention/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep technical analysis of LLM attention mechanisms, AttnGCG attacks, and how to detect semantic hijacking at the tensor level using mechanistic interpretability.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: MCP and the Expansion of the AI Attack Surface</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/mcp-model-context-protocol-security/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/mcp-model-context-protocol-security/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep architectural analysis of the Model Context Protocol (MCP). Understand how discovery layers, remote tool invocation, and malicious MCP registries transform isolated LLMs into vulnerable distributed systems.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Is MCP a Major New Attack Surface for AI Agents? Hybrid Analysis with MTGuard</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/mcp-security-hybrid-analysis-mtguard/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/mcp-security-hybrid-analysis-mtguard/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth technical breakdown of arXiv:2607.25297: evaluating Model Context Protocol (MCP) attack surfaces and how MTGuard combines static and dynamic co-analysis to defeat tool execution hijacking.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Architecture: Rethinking Agent Security as a Networking Problem</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/network-centric-agent-security/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/network-centric-agent-security/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A comprehensive architectural analysis of arXiv:2608.12172 (Tran et al., 2026). Discover how centralized control planes, distributed sidecar gateways, capability-based access, and Contextual Integrity redefine autonomous agent security.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: The OWASPification of Agentic AI</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/owaspification-agentic-ai/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/owaspification-agentic-ai/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Mapping classical cybersecurity concepts to modern AI exploitation. Understand how Prompt Injection is the new XSS, Tool Injection is the new RCE, and why Agentic AI is just the next iteration of AppSec.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Peer Pressure Works on AI Agents: Behavioral Contagion &amp; Consensus Hijacking in Swarms</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/peer-pressure-ai-agent-swarms-behavioral-contagion/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/peer-pressure-ai-agent-swarms-behavioral-contagion/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth technical analysis of Check Point Research (October 2026) demonstrating that multi-agent LLM swarms suffer from peer pressure. Observing non-compliant peers triggers a 44% to 98% rule-breaking escalation, bypassing alignment through implicit behavioral contagion.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Can One AI Hack Another AI? PIMiner and the Rise of Autonomous Prompt Injection</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/piminer-agent-against-agent-red-teaming/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/piminer-agent-against-agent-red-teaming/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Empirical breakdown of PIMiner (arXiv:2608.05108): how an autonomous red-teaming agent constructs transferable strategy libraries to break frontier AI defenses in under 10 queries.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] The Landscape of Prompt Injection in LLM Agents &amp; The AgentPI Trilemma (arXiv:2602.10453)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/prompt-injection-landscape-agentpi-benchmark/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/prompt-injection-landscape-agentpi-benchmark/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Comprehensive SoK breakdown of arXiv:2602.10453: context-dependent agent tasks, the fundamental Trustworthiness-Utility-Latency trilemma, the AgentPI benchmark, and 5 operational exploit patterns.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Prompt Injection vs Tool Poisoning vs RAG Poisoning vs MCP Attacks: A Comparative Attack Taxonomy</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/prompt-injection-vs-tool-poisoning-vs-rag-vs-mcp/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/prompt-injection-vs-tool-poisoning-vs-rag-vs-mcp/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Comprehensive comparative taxonomy dissecting the four major agentic attack vectors: execution layers, attacker prerequisites, blast radii, forensic artifacts, and defense boundaries.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: RAG Poisoning and Knowledge Base Manipulation</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/rag-poisoning/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/rag-poisoning/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Technical analysis of Retrieval-Augmented Generation (RAG) Poisoning, exploring semantic camouflage, multi-modal vulnerabilities (PoisonedEye), and attack vectors on vector databases.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Runtime Security &amp; Detection for AI Agents</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/runtime-security-ai-agents/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/runtime-security-ai-agents/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep architectural guide to runtime defense mechanisms for tool-enabled AI agents. Explore execution tracing, semantic drift detection, and the emergence of AI EDR.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Toward Secure LLM Agents: Threat Surfaces, Attacks, Defenses, and Evaluation (arXiv:2606.10749)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/secure-llm-agents-threat-surfaces-defenses-survey/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/secure-llm-agents-threat-surfaces-defenses-survey/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth deconstruction of the Nanjing University meta-survey (arXiv:2606.10749) analyzing 247 papers on autonomous LLM agent security: lifecycle threat modeling, weak defense compositionality, and the state corruption frontier.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Semantic Execution Layers &amp; Probabilistic Interpreters</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/semantic-execution-layers/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/semantic-execution-layers/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A profound architectural analysis of Agentic AI. Discover how LLMs act as probabilistic interpreters, translating natural language into deterministic system actions via semantic syscalls.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Can AI Agents Really Perform Reverse Engineering? SRE-Bench Reveals Their Limits</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/sre-bench-reverse-engineering-limits/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/sre-bench-reverse-engineering-limits/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth technical breakdown of SRE-Bench (arXiv:2608.11469): empirical evaluation of AI agents across 19 realistic binaries, 44 anti-analysis primitives, and 1,572 tasks.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] SysEvolve: Autonomous Attack-Defense Co-Evolution in Cyber Ranges — Analyzing arXiv:2608.15012</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/sysevolve-autonomous-cyber-co-evolution/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/sysevolve-autonomous-cyber-co-evolution/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Empirical analysis of arXiv:2608.15012: closed-loop co-evolution between autonomous offensive agents (SYSSPEAR) and defensive agents (SYSARMOR) across 1,148 multi-host ranges and 257 CVEs.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Threat-Preserving Representation Sensitivity (TPRS): Why Tool Naming Skews Security Benchmarks (arXiv:2610.03585)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/threat-preserving-representation-sensitivity-tprs/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/threat-preserving-representation-sensitivity-tprs/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>In-depth critique of Penn State&apos;s landmark study (arXiv:2610.03585): how semantic tool naming artificially swings agent Attack Success Rates by 11–13% across ASB, MCPTox, and AgentDojo.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: The Tokenization Layer and BPE Exploitation</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/tokenization-layer-exploitation/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/tokenization-layer-exploitation/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep technical analysis of how LLMs &apos;read&apos; text. Understand Byte Pair Encoding (BPE), Adversarial Tokenization, and how Threat Actors weaponize Glitch Tokens to bypass AI safety filters.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Tool Injection as the Convergence Layer of LLM Exploitation</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/tool-injection-architecture/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/tool-injection-architecture/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A definitive architectural and forensic analysis of Tool Injection in Agentic AI. Understanding how probabilistic interpreters translate semantic manipulation into operational exploitation.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Tool Poisoning &amp; The Semantic Supply Chain</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/tool-poisoning-semantic-supply-chain/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/tool-poisoning-semantic-supply-chain/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep architectural analysis of Tool Poisoning in Agentic AI ecosystems. Discover how malicious tool manifests, compromised MCP registries, and metadata poisoning hijack LLM orchestration.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Training Data Poisoning &amp; Backdoors</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/training-data-poisoning/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/training-data-poisoning/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A deep technical analysis of Training Data Poisoning in Large Language Models, exploring backdoor triggers, instruction-tuning vulnerabilities, and supply chain forensics.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] AI Security Research: Trust Boundary Collapse in Agentic AI</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/trust-boundary-collapse/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/trust-boundary-collapse/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A profound architectural analysis of the Trust Boundary Collapse in LLMs. Understand how instruction/data conflation and semantic ambiguity act as the root cause of all modern AI exploitation.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] Agentic LLM Vulnerabilities: The 4-Layer Taxonomy &amp; Systematic Review (arXiv:2608.10530)</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/vulnerabilities-agentic-llms-survey/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/vulnerabilities-agentic-llms-survey/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>A comprehensive architectural analysis of arXiv:2608.10530 (Hossain et al., 2026). Explore the first PRISMA systematic review on Agentic LLM security, the 4-layer taxonomy, and the critical 14-fold perception-action risk disparity.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
  <entry>
    <title>[AI Security Research] What Can AI Agents Actually Do in Cybersecurity in 2026? A Systematic Mapping of Capabilities, Benchmarks, and Limits</title>
    <link href="https://hermes-codex.vercel.app/ai-security/core-studies/what-can-ai-agents-do-cybersecurity-2026/"/>
    <id>https://hermes-codex.vercel.app/ai-security/core-studies/what-can-ai-agents-do-cybersecurity-2026/</id>
    <updated>2026-10-09T08:00:00Z</updated>
    <summary>Comprehensive meta-analysis and systematic mapping of LLM-based agents in software and systems security based on arXiv:2608.28490, synthesizing empirical data across 11 security domains, benchmark ecosystems, and structural capability limits.</summary>
    <category term="Agentic AI Security Research"/>
  </entry>
</feed>