Skip to content
Exploitation of Claude API Keys Enables Covert Remote Control Attacks

Exploitation of Claude API Keys Enables Covert Remote Control Attacks

First seen 12 Nov 2026, 14:17 UTC • •
API

Article Content

Browse articles
ThreatCluster AI
ThreatCluster •November 12, 2026 at 14:18 UTC
  • •Compromised Claude API keys can turn Claude Code agents into remote-controlled implants.
  • •The attack method exploits the memory stores feature of the Claude platform without needing vulnerabilities.
  • •Darktrace's tests demonstrated data exfiltration capabilities using legitimate traffic patterns.

Darktrace researchers demonstrated that compromised Claude API keys can turn Claude Code agents into remote-controlled implants. This method involves leveraging the memory stores feature of Anthropic’s Claude platform, allowing attackers to execute commands and exfiltrate data without relying on vulnerabilities in the platform. The attack requires initial access to a developer's machine to obtain user credentials. During testing, the researchers successfully pulled a 1 GB file from a simulated corporate environment. The technique mimics legitimate traffic, making detection challenging. Although the method has been tested in a lab, it has not been reported as used in the wild. The findings highlight the risks associated with AI tools in enterprise environments, particularly as organizations increasingly adopt AI capabilities.

Start a free Starter trial for enhanced analysis

Ask AI about this cluster

Updated 2d ago How this analysis works

Timeline

2026-10-21
CVE-2026-109618 published
A critical vulnerability was disclosed, affecting systems with a CVSS score of 9.2.
Helpnetsecurity
2026-10-21
CVE-2026-109619 published
Another critical vulnerability was disclosed, also with a CVSS score of 9.2.
Helpnetsecurity
2026-10-26
CVE-2026-109618 added to CISA KEV
CISA added this critical vulnerability to its Known Exploited Vulnerabilities catalog.
Helpnetsecurity

More articles in this cluster (5)

Following this threat?

Track Dustmaker and CVE-2026-109618 in your own feed — you're alerted when they show up in new reporting, leak sites or exploitation.

Free account · no card needed

Extracted Entities (20)

1 / 4

Common questions

What are the prerequisites for this attack?
An attacker must first compromise a developer's machine and obtain the user's Claude credentials.
Is this technique currently being exploited in the wild?
No, the technique has been demonstrated in a lab but has not been reported as used in real-world attacks.
What should organizations do to mitigate this risk?
Organizations should secure API keys and monitor for unusual activity involving AI tools within their environments.