This appendix gives the full code review behind the short summary in the main report. The FBI names this developer as the one who released ARTEX on GitHub in July 2026. The developer denies involvement in the attacks that used ARTEX.
This review describes what the code contains, not who has used it. ARTEX itself went closed-source before we could obtain it, so what we report about ARTEX comes from the firms that analyzed it. What we could obtain is everything else its developer has published, which comes to twenty public repositories under the GitHub account Autumn-27.
We preserved them as archived git histories in an isolated environment and examined the code directly, without building or running any of it.
ScopeSentry, a recon network rewired for AI agents
ScopeSentry is the developer's flagship project. The two-year-old platform (first commit June 2024, roughly 385 commits since) maps an organization's internet-facing footprint and hunts for weaknesses across a fleet of distributed scan nodes.
The nodes bundle the standard open-source offensive toolchain (nuclei, httpx, subfinder) along with ready-made programs ("binaries") that can't be checked against published source code, including a closed-source web crawler.
On June 10, 2026, six weeks before ARTEX appeared, the developer added a Model Context Protocol (MCP) server, a standard connector that lets AI assistants use other software, and a usage guide written for AI assistants. The server gives an AI agent 17 tools.
With them, the agent can create projects, launch distributed scan tasks against any list of targets and pull back what the scans find, from vulnerabilities and leaked sensitive data to subdomain-takeover candidates. The guide also lays out a two-phase reconnaissance workflow for the agent to follow. We found no hidden tools beyond the documented 17, and the AI interface cannot install new code.
Plugins that install without signature checks
ScopeSentry's plugins are raw Go source code. An embedded interpreter executes them inside the running server with full access to the host, including the ability to run system commands. Administrators can install them from a plugin store on the developer's own server, api.scope-sentry.top, which also lists paid plugins.
We found no cryptographic signature check on that code. Installed plugins re-run automatically at every startup and propagate to every scan node.
We found no malicious plugin.
Without signature checks, whoever controls that server controls the code every ScopeSentry deployment runs when it installs or updates plugins from the store.
Norma, an agent engine modeled on Claude Code
On July 25, 2026, one day before ARTEX's release, the developer published the first code of a project called Norma. It is a model-agnostic agent harness, written in Go, that turns any large language model into an autonomous agent with shell and file tools, sub-agents, memory and MCP support.
In 28 places, comments in its source tie Norma's settings and mechanisms to named internals of Anthropic's Claude Code, in one case describing them as "aligned 1:1 with claude-code." Those details were exposed, according to published reports, when Claude Code's source was reconstructed from source maps shipped in its npm package.
GitHub records show the same account forked that reconstructed source on March 31, 2026. The comments show that mapping of settings; on their own, they do not show that Norma copies Claude Code's code.
A Norma test still passes a setting (an "environment variable") named ARTEX_MARK through to its tools. That leftover points to shared code lineage with ARTEX, which comes from the same developer. It is an inference, not proof.
Norma's public examples are benign (a coding assistant and a read-only code auditor), and it was still being developed in public on October 18, ten days after its author said ARTEX would receive no further updates. That statement was about ARTEX, so continued work on a separate project does not contradict it.
What it adds up to
An agent engine, an AI control interface and a distributed scanning network are the components an autonomous reconnaissance agent would need. All three remain public, from the same developer, after ARTEX went closed. They sit in projects separate from ARTEX, the tool the developer's Oct 8 statement addressed.
Each piece is dual-use and has legitimate defensive value on its own. None of what we examined is the attack tool itself, and we found nothing in it that refers to the November outages.
Key dates in the developer's public code
These dates come from the repositories' own GitHub records unless another source is named. They describe when code was published, not how anyone used it.
- June 5, 2024: first commit to ScopeSentry.
- March 31, 2026: the account forks the reconstructed Claude Code source.
- June 10, 2026: ScopeSentry gains its MCP server and AI usage guide, which shipped in release 1.9.2 on June 12.
- July 25, 2026: first code published in Norma.
- July 26, 2026: ARTEX published on GitHub, according to Censys.
- October 8, 2026: the developer's statement; the FBI says the ARTEX repository was made unavailable that day, following reports of abuse.
- October 18, 2026: latest public update to Norma, as of our check on November 17.
A note on method: we found no hidden or AI-directed prompt-injection text in any repository; the only invisible characters were standard emoji modifiers. A November 17 commit that mentions ARTEX is an unmerged pull request from a different account, not the developer's. Git timezone stamps (China Standard Time on two years of ScopeSentry commits, US Eastern on most Norma commits) reflect machine settings and are not evidence of location.
Sources for this appendix
- Autumn-27 GitHub profile, including the Oct 8, 2026 ARTEX statement (primary source)
- ScopeSentry source: MCP server and plugin system (primary source, examined by Recent Breaches)
- Norma agent harness source (primary source, examined by Recent Breaches)
- ARTEX: Agentic Asbestos (Censys)
- FLASH-20261113-001: Threat Overview of ARTEX Autonomous Penetration Testing Framework (FBI / IC3)
- Claude Code's source code appears to have leaked: here's what we know (VentureBeat)
Update log: This is a developing story. We'll add new findings here as the FAA forensics, vendor advisories and official root-cause reports come in. 18 Nov 2026: added "Where Anthropic's Claude and Mythos 5 fit", including whether restricting Mythos helped or hurt, the publisher's view and a skeptical reading, a key point, 21 timeline entries (AI models, the June 2026 US order, ARTEX's developer, related attacks, and the November 16 Optimizely and Coles incidents), and an item to "What would settle it". Later on 18 Nov: added the Director of National Intelligence's statements and House testimony, the FBI's clarification on ARTEX, the FAA's router pause and the state of its investigation, the Secretary of State's remarks, an unconfirmed report that fixes for AI-found flaws caused the outages, updates from Equinor and Optimizely, Visa's statement, earlier Amazon, Google and StepSecurity reports on ARTEX, a GAO report on AI chatbots, and 14 timeline entries. Last updated 18 Nov 2026.