0x78129b657812…78129b62

ConfirmedSecurity591 vB167 sat/vB3 min decode

Google Ties Crypto Firm Intrusions to Suspected North Korean Hackers

GTIG's Nov. 16 update links a crypto-sector intrusion campaign to suspected North Korean hackers using a Claude Code backdoor and an autonomous exploit framework.

Google Links Crypto Campaign to Suspected North Korean Hackers
WitnessGoogle Links Crypto Campaign to Suspected North Korean Hackers377053 / pixabay

Outputs

  1. GTIG published its quarterly AI threat update on Nov. 16, attributing a crypto-sector campaign to suspected North Korean cluster MIDNIGHT NEPTUNE.

  2. Altered .claude/settings.json files triggered download of the SOMBERMEME backdoor when developers opened Claude Code.

  3. Zero-day exploits averaged 11 per month from January to August 2026, up from 8 per month in 2025, peaking at 22 in August.

  4. The exploit_agent framework on an exposed staging server drafted and executed exploit code without human intervention.

  5. Attackers used open-source AI pen-testing frameworks CAIRN and PentAGI in observed intrusions.

Google's Threat Intelligence Group (GTIG) has attributed a campaign against cryptocurrency companies in North America and Europe to suspected North Korean hackers who altered Claude Code settings to deliver a backdoor onto developer machines. The findings appeared in GTIG's quarterly AI threat update, published Nov. 16.

GTIG tracks the suspected cluster under the name MIDNIGHT NEPTUNE. The operators posed as recruiters through fake profiles on professional networking platforms, then steered targets toward fraudulent video meeting platforms and pressed them to run diagnostic-repair scripts. GTIG assessed those scripts as likely written with large language models. In plain words, the attackers appear to have used AI to write the very bait that opened the door.

How did the attackers move inside victim networks?

After gaining initial access, the operators modified .claude/settings.json files inside victims' internal code repositories. The change was mechanical and effective: when a developer opened Claude Code, the altered configuration triggered a download of the SOMBERMEME backdoor. The intrusion therefore did not depend on a novel exploit, only on a trusted file that developers rarely inspect.

Persistence followed a second, quieter step. The operators created self-generated GitHub personal access tokens and deliberately named them to resemble Microsoft Copilot, disguising their movement through company systems. GTIG did not state how many firms were compromised or how long the campaign ran before detection — a gap that limits any estimate of the damage.

What is exploit_agent?

A separate finding in the same update concerns a framework GTIG calls exploit_agent, discovered on an exposed staging server. The framework linked frontier and local AI models and was configured to compare software code before and after patches, identify possible vulnerabilities, draft exploit code and execute it without human intervention.

That last phrase deserves emphasis. A tool that drafts and runs exploits autonomously removes the human bottleneck that traditionally constrained offensive operations, even if automation can introduce detectable noise.

The server also held stolen data alongside scripts for reconnaissance, brute-force attempts and the installation of web backdoors. Its telemetry suggested apparent interest in French websites focused on media, entertainment and education. One example involved access-control code in version 2.19.25 of Spectra (Ultimate Addons for Gutenberg), a WordPress plugin. GTIG did not say whether that specific path led to a confirmed breach.

Are disclosed exploits accelerating?

GTIG's analysis found that vulnerability disclosures paired with exploits nearly doubled between January and August 2026. Zero-day exploits averaged 11 per month during that period, up from eight per month in 2025, and reached 22 in August alone. Put another way, defenders now face roughly one new zero-day exploit every 34 hours, against one every three and a half days in the prior year.

The report also noted that attackers deployed open-source AI penetration-testing frameworks, including CAIRN and PentAGI, in observed intrusions. These tools can compress attack activity to machine speed. That speed carries a cost for the attacker: reduced stealth can make intrusions easier to detect. Whether defenders can act on that window fast enough remains the operative question.

What should crypto firms take from this?

For crypto companies, the operational consequence is direct. Developer tooling — configuration files in internal repositories, tokens named after familiar services — has become an attack surface in its own right. Firms that audit .claude/settings.json and inventory GitHub personal access tokens will close at least two of the doors MIDNIGHT NEPTUNE used. Those that do not should expect GTIG's next quarterly update to describe similar paths, likely at greater scale.

via tokenpost.com (Source)

More from Elena Vasquez

Elena Vasquez

Show full bio

Staff writer covering regulation, policy and their impact on crypto and web3 industry at Mempool Brief.

1,525 articles