SpecterOps Releases Blacklight for AI Agent Endpoint Artifacts

The project focuses on local files that may be accessible after an attacker compromises a user's account or workstation. SpecterOps says authentication stores can contain access or refresh tokens, account identifiers, and session metadata. Configuration files may describe trusted projects, sandbox rules, approval settings, environment variables, and Model Context Protocol servers, while transcripts and activity indexes can expose repository names, internal paths, deployment procedures, and the user's recent work.

That distinction matters: this is not a newly disclosed remote vulnerability in Codex, Claude Code, Cursor, or Antigravity. The risk comes from sensitive application state already present within the compromised user's security context. Blacklight is meant to help authorized operators and defenders inventory that state and decide what warrants closer review.

Blacklight Scout includes loaders for the three major desktop operating systems. The quieter loaders identify agent roots, file metadata, and a small ranked set of recent session candidates without reading artifact bodies. Windows executables can perform deeper metadata analysis, but the project says they are designed not to print credential values, identities, endpoints, or session text.

When deeper analysis is justified, selected artifacts can be processed offline. The session-analysis command recognizes supported formats, produces redacted indicators, and ranks reports for follow-up. The repository also includes osquery configurations and SIEM guidance for defenders. SpecterOps specifically notes that Windows read detection can use Security Event 4663 with Audit File System enabled and appropriate SACLs, while macOS and Linux coverage depends on their available file-event telemetry.

AI agent profiles are becoming another high-value endpoint data store, alongside browser profiles, shell history, and cloud CLI configuration. For practitioners, the immediate task is not blanket transcript collection. It is to inventory which agents are installed, restrict access to their local state, define retention rules for sessions, and monitor sensitive credential paths with controls appropriate to each operating system. Blacklight turns that assessment into a repeatable workflow while keeping collection selective.