Collection Scope

User request: collect latest AI security news, blogs, white papers, papers, and technical documents.

This batch prioritizes sources not already present in raw/ or 01_Sources/, with emphasis on:

• 2026-06-19 or newer sources.
• Official and primary technical sources before secondary news.
• Agentic AI, MCP, multi-agent systems, skill supply chain, prompt injection in the wild, AI for Security, and industry trend reports.

Saved Sources

Primary / technical

1. raw/papers/mcp-privacy-leakage-risks-2026.md

• MCP privacy leakage static analysis; submitted 2026-06-19.
• Relevance: local-to-LLM privacy boundary, credential/API key exposure, MCP server vetting.

1. raw/clips/unit42-openclaw-skill-marketplace-supply-chain-2026.md

• Unit 42 OpenClaw/ClawHub skill-marketplace threat research; published 2026-06-23.
• Relevance: malicious skills, scanner evasion, agent supply-chain integrity.

1. raw/clips/google-prompt-injections-web-2026.md

• Google analysis of prompt injection patterns on the public web; published 2026-04-23.
• Relevance: observed IPI in the wild and operational monitoring approach.

1. raw/papers/agent-communication-graph-metadata-2026.md

• Communication graph metadata leakage in A2A/MCP-style autonomous agent workflows.

1. raw/papers/gt-mcp-contextual-reasoning-2026.md

• Context poisoning defense using controller-driven multi-agent MCP and rollback.

1. raw/papers/openagenet-oan-trust-governed-resource-identity-2026.md

• Trust-governed resource identity and discovery for agent interconnection.

1. raw/papers/ss-zkr-multi-agent-routing-2026.md

• Privacy-preserving semantic routing for multi-agent collaboration.

1. raw/papers/rails-agentic-commerce-clearing-2026.md

• Verification-native clearing for financially material agentic commerce.

1. raw/papers/agents-all-the-way-down-2026.md

• Agent methodology source for AI technology attack-surface mapping.

Industry reports / white papers

1. raw/whitepapers/cisco-state-ai-security-2026.md
2. raw/whitepapers/google-cloud-cybersecurity-forecast-2026.md
3. raw/whitepapers/wef-global-cybersecurity-outlook-2026.md
4. raw/whitepapers/darktrace-state-ai-cybersecurity-2026.md
5. raw/whitepapers/paloalto-securing-agentic-ai-2026.md
6. raw/clips/salt-ai-api-security-report-2026.md

News / secondary trend coverage

1. raw/news/darkreading-openclaw-malicious-skills-2026-06-25.md
2. raw/news/axios-global-ai-wars-2026-06-24.md

Skipped As Duplicates

• Microsoft AutoJack, 2026-06-18: already present as 01_Sources/microsoft-autojack-2026 - AutoJack.md and raw/news/microsoft-autojack-2026-06-19.md.
• Microsoft Claude Code GitHub Action case, 2026-06-05: already present as source and raw clip.
• OWASP State of Agentic AI Security and Governance 2.01: already present as source and raw whitepaper.
• Several June 2026 arXiv agent security papers were already present, including Toward Secure LLM Agents, VATS, SecureClaw, ZERO-APT, Stored Prompt Injection, RAG Inference Cost Attacks, and SkillVetBench.

Recommended Ingest Order

1. MCP and tool-surface risks:

• mcp-privacy-leakage-risks-2026
• gt-mcp-contextual-reasoning-2026
• agent-communication-graph-metadata-2026

1. Agent supply chain and skill ecosystems:

• unit42-openclaw-skill-marketplace-supply-chain-2026
• darkreading-openclaw-malicious-skills-2026-06-25
• agents-all-the-way-down-2026

1. Agent identity, routing, and delegated-action assurance:

• openagenet-oan-trust-governed-resource-identity-2026
• ss-zkr-multi-agent-routing-2026
• rails-agentic-commerce-clearing-2026

1. Macro trend and industry reports:

• google-prompt-injections-web-2026
• google-cloud-cybersecurity-forecast-2026
• cisco-state-ai-security-2026
• paloalto-securing-agentic-ai-2026
• wef-global-cybersecurity-outlook-2026
• darktrace-state-ai-cybersecurity-2026
• salt-ai-api-security-report-2026
• axios-global-ai-wars-2026-06-24

Fetch Notes

• raw/news/darkreading-openclaw-malicious-skills-2026-06-25.md: direct PowerShell Invoke-WebRequest returned HTTP 403 during verification, but search metadata and summary were available. Use as secondary news only.
• raw/news/axios-global-ai-wars-2026-06-24.md: web fetch succeeded and exposed enough metadata for source capture.

Research Areas Affected

• MCP server privacy leakage and implicit exfiltration.
• Prompt injection in the wild and public-web monitoring.
• Agent skill marketplace trust, scanner evasion, and local executor risk.
• Multi-agent workflow metadata leakage and workflow integrity.
• Agent resource identity, discovery, routing privacy, and delegated-action clearing.
• AI for Security: vulnerability discovery, SOC agent adoption, and benchmark-to-operation gaps.
• Frontier AI cyber capability diffusion and international governance pressure.