Google reports AI-tool tampering in attacks on crypto firms
Google described intrusions at crypto firms involving fake recruiters and tampered Claude Code settings used to download a backdoor after initial access.

Google Threat Intelligence Group described intrusions at cryptocurrency organizations in North America and Europe in its November 16 report. After obtaining initial access, attackers changed Claude Code settings in internal repositories so that starting the tool triggered a download of the SOMBERMEME backdoor. Google tentatively linked the activity to a suspected MIDNIGHT NEPTUNE cluster in its original report.
Details
The campaign began with fraudulent recruiter profiles and invitations to fake meeting platforms. Google said profile images contained SynthID watermarks indicating AI generation. Targets were then persuaded to run supposed diagnostic fixes. Researchers described this stage as ClickFix social engineering, rather than an attack that independently broke a blockchain's cryptography.
After compromise, the attackers moved into internal repositories and altered the coding tool's configuration. The AI product became part of access they had already gained to the working environment. CryptoSlate reported this part of the findings on November 17. The public account does not identify the affected cryptocurrency companies or establish an amount stolen from them.
The report's attribution is qualified: Google refers to a suspected MIDNIGHT NEPTUNE cluster. That should be distinguished from conclusively identifying individual perpetrators. Similarly, evidence of generated profile photographs does not establish that every step in the campaign was carried out by autonomous AI.
What it means for markets
For cryptocurrency companies, the exposure involves development access and internal systems. Routine configuration can determine what runs in an employee's environment. Protecting keys and infrastructure therefore depends on account security and control over repository changes, as well as the correctness of smart-contract code.
Google's broader report also discusses separate attack-automation frameworks. Those observations should not be combined with this campaign to claim that an entirely autonomous system stole cryptocurrency. What is documented here is abuse of workplace AI tooling following an intrusion; no public loss calculation is provided for this campaign. Readers can find wallet-security fundamentals in the seed-phrase guide and the wider market backdrop in the daily brief.



Comments