What happened
Cryptocurrency organizations / Google reported observations across multiple intrusions.
- 1
Fraudulent recruiting meetings reached employees
Initial access across the reported intrusionsPeople posing as recruiters brought targets into fraudulent meetings and persuaded them to run software on employee devices.
- 2
Internal repository settings were changed
After initial compromise in the reported activityGTIG reported unauthorized changes to internal repositories that affected developers’ coding tools. It did not name the code-hosting provider or identify the credential used for the changes.
- 3
Repository changes could trigger a backdoor download
During the reported intrusionsThe changes configured Claude Code to download SOMBERMEME on developer machines. GTIG did not identify a completed download or execution on an individual endpoint.
- 4
GitHub personal access tokens were also created
Separately observed in the campaignGTIG also observed the creation of GitHub personal access tokens. The report does not identify their types or permissions, or show that they were used to change an internal repository.
Impact
Attackers changed internal repositories so that Claude Code was configured to download the SOMBERMEME backdoor on developer machines. GTIG also observed GitHub personal access tokens being created. It did not report a loss total or establish what the tokens accessed.
Could this have been prevented?
Restrict project-defined automation on managed developer machines
On managed Claude Code clients, block hooks supplied by projects. Keep the automations your administrators and development tools are allowed to run under separate review. GitHub’s token settings govern access to organization resources after a token exists.
Incident control assessments
- 0
- Prevented
- 0
- Contained
- 4
- Not covered
How it could have been prevented
The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.
Cryptocurrency organizations / Google reported observations across multiple intrusions.
1 · Fraudulent recruiting meetings reached employees
1 Not covered2 · Internal repository settings were changed
1 Not covered3 · Repository changes could trigger a backdoor download
1 Not covered4 · GitHub personal access tokens were also created
1 Not coveredKey facts
- GTIG attributed these intrusions to a suspected MIDNIGHT NEPTUNE cluster. It discussed ARTEX separately in the same report.
- GTIG reported multiple intrusions at unnamed cryptocurrency organizations in North America and Europe. It did not give a victim count or dates for individual cases.
- The investigators linked deceptive recruiting personas to synthetic profile images. Targets were persuaded to run software during purported meetings.
- GTIG found internal repository changes that could prompt developer systems to download SOMBERMEME. It did not identify the repository host or the branches affected.
- The attackers created GitHub personal access tokens. GTIG did not identify their types, permissions or any repository operation performed with them.
Sources
Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.
- 1.GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant ThresholdGoogle Threat Intelligence Group · Nov 16, 2026 · ResearchGTIG observations of suspected MIDNIGHT NEPTUNE activity, repository changes, a configured SOMBERMEME download and GitHub token creation.
- 2.Setting a personal access token policy for your organizationGitHub · undated · AdvisoryOrganization policy governs token access to resources the organization owns.
- 3.All settingsAnthropic · undated · AdvisoryManaged Claude Code policy can exclude project-defined hooks on enrolled developer machines.
The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.