SecurityWeek’s weekly cybersecurity news roundup provides a comprehensive overview of the most significant developments across the digital threat landscape. As the cybersecurity environment becomes increasingly complex, staying informed about emerging attack vectors, policy shifts, and industry vulnerabilities is essential for security professionals and organizational leaders. This week’s report covers a broad spectrum of incidents, ranging from sophisticated AI-focused malware campaigns to critical infrastructure concerns and major legal developments in the fight against dark web marketplaces.
PoeLLM Malware Leverages GitHub for Stealthy Command-and-Control
A notable development in the threat landscape involves "PoeLLM," a malware strain identified by Black Lotus Labs that has been active since at least April 2026. This campaign specifically targets exposed artificial intelligence and open-source infrastructure, with a particular focus on services such as LiteLLM, Ollama, Gotenberg, and Gitea. The primary objective of the attackers appears to be cryptocurrency mining and the expansion of a growing botnet.
What makes PoeLLM particularly insidious is its evasion technique. Rather than hardcoding a Command-and-Control (C&C) server address—which would be easily detected by static analysis—the malware retrieves its instructions from a poem hosted on GitHub. Infected machines are programmed to extract four specific keywords from the poem, which are then algorithmically converted into the IP address of the active C&C server. This method allows the threat actor to pivot to a new infrastructure by simply updating the text of the poem. Researchers have observed the poem being updated 11 times, suggesting an active and responsive operator, assessed to be an Italian speaker. This technique highlights the evolving nature of malware persistence, where attackers are increasingly leveraging legitimate, high-reputation services to bypass security filters.
GhostAction Campaign Continues to Target GitHub Secrets
The supply chain threat remains a significant concern for developers, as evidenced by the resurgence of the "GhostAction" campaign. According to data from GitGuardian, this secret-stealing operation successfully pushed its malicious GitHub Actions workflow to 772 public repositories between August 31 and September 30. The campaign targeted a total of 2,577 sensitive secrets, including SSH keys, Azure credentials, AWS access keys, and various database credentials.
GitGuardian’s analysis indicates that the campaign never truly ceased. Despite some minor changes, such as the implementation of a new exfiltration server, the attackers largely relied on the same playbook observed in 2025. By embedding malicious actions into public repositories, the threat actors gain a foothold in development environments, potentially leading to broader breaches within organizations. This ongoing campaign serves as a stark reminder of the risks associated with the automated nature of modern CI/CD pipelines and the necessity for rigorous secret scanning and monitoring.
Justice Served: Uranium Finance Hacker Convicted
In the realm of decentralized finance, a federal jury has convicted 36-year-old Jonathan Spalletta of Maryland for his role in two separate hacks of the Uranium Finance crypto exchange. The legal proceedings detailed a systematic exploitation of smart contract vulnerabilities that allowed Spalletta to siphon approximately $1.4 million in the first attack and an additional $53.3 million in a subsequent breach. The latter incident was catastrophic enough to force the permanent shutdown of the exchange.
Following the thefts, Spalletta engaged in an elaborate money-laundering scheme, utilizing services like Tornado Cash to obfuscate the origin of the funds. The proceeds were then converted into physical assets, including rare Pokémon and Magic: The Gathering trading cards, as well as antique Roman coins. Spalletta now faces a significant prison sentence, with up to 10 years for computer fraud and up to 20 years for money laundering, marking a significant win for federal authorities tasked with policing the intersection of cryptocurrency and criminal activity.
CISA Updates Cybersecurity Retention Incentives
The Cybersecurity and Infrastructure Security Agency (CISA) has announced a tightening of its criteria for the Cybersecurity Retention Incentive program. While the agency will maintain the program—which offers incentives worth up to 25% of a staff member’s base salary—through the 2027 fiscal year, the path to qualification is becoming more stringent.
Under the new guidelines, staff must maintain an "exceeds expectations" performance rating and dedicate at least 51% of their professional time to cyber-related duties within specific job series. For those outside these designated series, the threshold for eligibility is higher, requiring at least 75% of their time to be focused on cybersecurity, subject to approval by a review board. This strategic overhaul comes in response to a Department of Homeland Security (DHS) inspector general report, which criticized the previous management of the program, noting that it had been applied too broadly and lacked the necessary oversight to ensure the retention of truly critical cyber personnel.
OTCC Proposes Security Directive for Federal Operational Technology
The Operational Technology Cybersecurity Coalition (OTCC) is pushing for increased focus on the security of Operational Technology (OT) within federal civilian agencies. In a recent proposal, the coalition outlined what a CISA Binding Operational Directive (BOD) should entail for agencies that manage over 8,000 facilities. These facilities house critical building automation systems, including HVAC, power management, and physical access control systems.
The OTCC recommends that any such directive require agencies to designate a senior official or specialized office with direct accountability for OT security. Furthermore, the proposal calls for the application of existing security standards and a dedicated effort to prioritize CISA’s Cybersecurity Performance Goals. By formalizing these requirements, the coalition hopes to address the significant security gaps that currently exist in federal building management systems.
Credential Stuffing Targets Domino’s Customers
Domino’s has initiated an account reset for a small group of customers following reports of unauthorized access via credential stuffing. The company confirmed that its internal systems were not compromised, but rather that attackers used email and password combinations leaked in unrelated third-party data breaches to gain entry to customer accounts. Domino’s, which does not store payment information, moved quickly to reset the affected accounts and notify users, underscoring the persistent threat of credential reuse in an era of frequent data breaches.
AI Suspected in South Korean Bank Attacks
In South Korea, government officials and investigators are probing a series of cyberattacks against the nation’s banking sector, with President Lee Jae Myung noting credible signs that AI tools were utilized in the orchestration of the breaches. While authorities have not disclosed the specific AI technologies involved, the investigation has revealed that sensitive customer personal information was compromised.
CrowdStrike has added weight to these suspicions, reporting the discovery of ARTEX configuration files, Claude memory files, and Claude Code session histories within the attackers’ infrastructure. The security firm has expressed moderate confidence that the attacks were carried out by a Chinese-speaking threat actor driven by financial motives. This incident represents a growing trend of adversaries weaponizing generative AI to automate reconnaissance and exploit development.
Empire Market Co-founder Receives 40-Year Sentence
The dark web marketplace ecosystem suffered another blow as Raheim Hamilton, a 30-year-old from Virginia, was sentenced to 40 years in federal prison. Hamilton was one of the co-creators of Empire Market, a platform that operated between 2018 and 2020 and facilitated over four million transactions worth more than $430 million. The market was a hub for illicit activity, ranging from the sale of narcotics and counterfeit currency to the distribution of stolen credentials and sophisticated hacking tools. Hamilton’s partner, Thomas Pavey, previously pleaded guilty and is expected to be sentenced later this month.
Critical Vulnerability in Nvidia GPU Monitoring Tools
Security researchers have disclosed a high-severity vulnerability, tracked as CVE-2026-47483, affecting Nvidia’s DCGM Exporter. This tool, which is used for monitoring GPU telemetry, contained a flaw that allowed unauthenticated attackers to flood profiling endpoints with requests, resulting in resource exhaustion and service crashes. This could lead to significant performance degradation for AI workloads hosted on the same hardware.
Between March and May 2026, researchers identified approximately 2,100 hosts that had the exporter exposed to the public internet without authentication, potentially exposing telemetry data from over 12,000 GPUs. Nvidia has released a fix, and organizations are strongly encouraged to update to version 4.8.2 or later to mitigate the risk of exploitation.
Tensorlake SDK Compromised in Supply Chain Attack
Finally, a supply chain compromise has affected the npm ecosystem. Version 0.5.144 of the tensorlake SDK—a tool used for AI agent sandboxes—was found to contain a credential-stealing worm. Security researchers from Socket and Sonatype described the incident as a "Shai-Hulud" style attack, where the malicious code harvests credentials for Kubernetes, AWS, GitHub, npm, and Vault, as well as configurations for various AI coding tools. The malware also possesses the capability to execute attacker-supplied code and propagate itself by republishing through other packages the victim may have authorization to publish. This incident serves as a critical warning for development teams to audit their dependencies and implement strict supply chain security practices.
