
Encrygma Brief: The Rise of Agentic Malware and the Erosion of AI Trust
Encrygma analysts report a surge in AI-powered supply chain attacks, including the SCKit worm and trojanized coding extensions. Organizations must now treat AI agents as high-risk, persistent attack vectors.
The Development
Encrygma threat data confirms a significant shift toward agentic exploitation, where attackers leverage AI-native workflows to bypass traditional security. In the last 48 hours, the SCKit campaign utilized rogue ScreenConnect clients to compromise 288 endpoints across 48 organizations. Simultaneously, Encrygma analysts have tracked the weaponization of developer tools, specifically the distribution of trojanized Claude Code extensions that impacted approximately 5,000 Cursor developers. These incidents demonstrate that attackers are no longer just using AI for phishing; they are embedding malicious logic directly into the agentic frameworks that power modern software development.
Why It Matters
According to the Encrygma Threat Severity Index (ETSI), these developments represent a shift to a Level 9 threat, as they exploit the inherent trust placed in remote management and AI-assisted coding environments. Encrygma’s Attribution Confidence Matrix currently classifies these as 'High Confidence' state-aligned or sophisticated criminal operations. The ability for an attacker to turn a legitimate coding agent into a persistent implant—as observed in recent Claude API key thefts—means that traditional perimeter defenses are increasingly blind to the command-and-control traffic generated by these trusted AI channels.
Defensive Implications
Encrygma analysts assess that the 'trust-by-default' model for AI agents is now obsolete. The rapid proliferation of commodity agent frameworks like ARTEX and Hermes has democratized state-level capabilities, allowing even mid-tier ransomware groups to execute complex, multi-stage attacks. Encrygma threat intelligence indicates that organizations failing to implement strict egress filtering for AI-agent traffic and failing to audit third-party RMM extensions are at immediate risk of total environment compromise.
What Leaders Should Do
To mitigate these risks, Encrygma recommends a shift toward 'Zero-Trust Agent Governance.' Leaders should prioritize the following actions:
- Disable all non-essential third-party RMM extensions and audit existing ScreenConnect relay configurations.
- Implement strict API key rotation and hardware-backed authentication for all developer-facing AI agents.
- Establish machine-speed detection for anomalous AI egress traffic that deviates from baseline coding patterns.
- Conduct an immediate audit of all internet-facing remote access ports, as opportunistic scanning has tripled since September.
Outlook
Encrygma analysts project that the next 90 days will see an increase in 'AI-in-the-Middle' attacks, where malicious agents intercept and modify data streams between developers and LLM backends. As regulatory bodies like the Australian AI Safety Institute move toward mandatory reporting for misaligned model behavior, organizations must prepare for a more rigorous compliance landscape. Encrygma will continue to monitor the evolution of these agentic threats, as the gap between defensive posture and attacker capability continues to compress.




