Blacklight Illuminating AI Agent Artifacts for Attackers and...

[carlospolop]

🤖 Automated Content Update

This PR was automatically generated by the HackTricks News Bot based on a technical blog post.

📝 Source Information

• Blog URL: https://specterops.io/blog/2026/08/12/blacklight-ai-agent-endpoint-artifacts
• Blog Title: Blacklight: Illuminating AI Agent Artifacts for Attackers and Defenders
• Suggested Section: AI > AI Security > AI MCP Security, with cross-references from Generic Methodologies & Resources > Phishing Methodology > AI Agent Abuse Local AI CLI Tools and MCP; proposed subsection: AI Agent Artifact Discovery and Post-Exploitation

🎯 Content Summary

Classification and impact: Blacklight is an open-source, cross-platform security research toolkit for discovering and analyzing local artifacts created by Codex, Claude Code, Cursor, and Antigravity/Gemini CLI. The project is intended for authorized assessments and is available from GitHub. It is post-compromise discovery and collection tradecraft rather than a CVE disclosure: the article does not identify vulnerable applica...

🔧 Technical Details

Known-root agent discovery: On an already accessed endpoint, enumerate a static catalog of per-user paths for installed AI agents and test only filesystem existence and type. Using platform-specific loaders such as a BOF, dylib, or shared object allows low-noise discovery of active agent roots without reading file contents.

Metadata-first post-exploitation triage: Inspect configuration, rules, authentication-file metadata, indexes, timestamps, sizes, and activity counts before collecting transcripts. Rank credential-related paths first and return a small number of newest, largest, or most recent session candidates. Keep token values, identities, endpoints, and chat text off the wire during initial triage, then require human review before downloading selected files.

Selective offline session mining: Transfer only selected agent files through the C2 channel and recursively parse them offline. A schema-aware parser can reco...

🤖 Agent Actions

Plan:

1. Inspect repository structure and at least five relevant AI/MCP/phishing pages.
2. Compare existing coverage with the supplied Blacklight research and verify primary-source repository details.
3. Select one best existing page, implement a concise non-duplicative section with numbered citations.
4. Re-read the edit, check links/style/reference placement, and review the final diff.
   The repository already has an AI/MCP page and a closely related local-agent-abuse page. I’ll compare those with adjacent AI, phishing, and post-exploitation pages before choosing a single edit location, as required by the non-duplication rule.
   The existing ai-agent-abuse-local-ai-cli-tools-and-mcp.md page is the strongest fit: it already documents local agent logs, credential discovery, and MCP trust boundaries. The new value is a post-compromise, metadata-first collection workflow—not another MCP vulnerability—so I’ll place it there rather than creating a CVE/tool-specific page or duplicating it across AI pages. Before editing, I’m finishing the required adjacent-page review and verifying Blacklight’s primary repository.
   Verification confirms a useful gap: the page only lists two log paths, while Blacklight adds a reproducible four-agent artifact catalog, low-noise metadata triage, schema-aware offline parsing, and OS-specific read-detection boundaries. I’ll add one concise section to that existing page and append the primary GitHub repository before the research blog in its references.
   The edit is in place in one file, with the existing references preserved and new sources appended as [14] and [15]. Next I’m checking citation integrity, forbidden-source absence, markdown whitespace, and whether the book build accepts the page.
   Updated:

src/generic-methodologies-and-resources/phishing-methodology/ai-agent-abuse-local-ai-cli-tools-and-mcp.md

Added:

• Cross-platform high-value artifact paths for Codex, Claude Code, Cursor, and Antigravity CLI.
• Metadata-first discovery and selective collection workflow.
• Blacklight offline analysis command and structured-store handling.
• Linux, macOS, and Windows artifact-access detection guidance.
• Numbered references to the Blacklight repository and SpecterOps research.

Validation passed for diff formatting, citation integrity, balanced code fences, forbidden sources, and source availability. mdbook build could not run because mdbook is not installed.

---

This PR was automatically created by the HackTricks Feed Bot. Please review the changes carefully before merging.

[carlospolop]

🔗 Additional Context

Original Blog Post: https://specterops.io/blog/2026/08/12/blacklight-ai-agent-endpoint-artifacts

Content Categories: Based on the analysis, this content was categorized under "AI > AI Security > AI MCP Security, with cross-references from Generic Methodologies & Resources > Phishing Methodology > AI Agent Abuse Local AI CLI Tools and MCP; proposed subsection: AI Agent Artifact Discovery and Post-Exploitation".

Repository Maintenance:

• MD Files Formatting: 1032 files processed

Review Notes:

• This content was automatically processed and may require human review for accuracy
• Check that the placement within the repository structure is appropriate
• Verify that all technical details are correct and up-to-date
• All .md files have been checked for proper formatting (headers, includes, etc.)

Bot Version: HackTricks News Bot v1.0