The global cybersecurity landscape continues to shift at a rapid pace, with threat actors increasingly targeting modern software development pipelines, decentralized finance platforms, and artificial intelligence infrastructure. SecurityWeek’s weekly cybersecurity news roundup provides a concise yet comprehensive overview of critical developments that may not receive full standalone coverage, yet remain vital to understanding the broader threat environment.
This curated summary spans vulnerability disclosures, emerging attack methodologies, federal policy updates, industry reports, and significant law enforcement actions designed to help security professionals, administrators, and stakeholders stay well-informed about the evolving risks facing digital ecosystems.
PoeLLM Malware Finds its C&C Server Hidden in a GitHub Poem
Among the most novel emerging threats detailed this week, security researchers at Black Lotus Labs have published findings on "PoeLLM," a sophisticated malware active since at least April 2026. The malware specifically targets exposed artificial intelligence and open-source services, with a primary focus on tools like LiteLLM, Ollama, Gotenberg, and Gitea. Once a machine is successfully infected, the malware utilizes the compromised environment for cryptocurrency mining and to expand a broader botnet.
What sets PoeLLM apart is its unique command and control (C&C) communication mechanism. Infected machines autonomously extract four distinct keywords from a poem hosted on a public GitHub repository. These keywords are subsequently translated into the IP address of the current C&C server. This architecture allows the threat operators to dynamically switch their infrastructure simply by modifying those specific words within the poem. According to telemetry, the operator—assessed by researchers to be Italian-speaking—has updated the underlying poem 11 times to shift command servers and evade blacklisting.
GhostAction Secret Theft Spreads to 772 More GitHub Repositories
Supply chain security remains a persistent headache for software engineering teams. Threat intelligence firm GitGuardian reported that the "GhostAction" supply chain campaign actively pushed its credential-harvesting GitHub Actions workflow to an additional 772 public repositories between August 31 and September 30. These repositories belong to 373 distinct users and organizations, putting roughly 2,577 sensitive secrets at risk. Among the harvested credentials are vital authentication assets, including SSH keys, database credentials, and access tokens for major cloud environments like Amazon Web Services and Microsoft Azure.
While the campaign utilized a newly established exfiltration server during this latest wave, researchers noted that the core playbook mirrors tactics observed throughout 2025. GitGuardian’s continuous monitoring data indicates that this campaign has operated with remarkable persistence, suggesting that automated scanning for misconfigured continuous integration and continuous deployment (CI/CD) pipelines remains a highly lucrative vector for automated attackers.
Jury Convicts Uranium Finance Hacker in Multi-Million Dollar Cryptocurrency Fraud
In the realm of law enforcement and cryptocurrency security, a federal jury has convicted 36-year-old Jonathan Spalletta of Maryland on multiple counts of computer fraud and money laundering. The conviction stems from two major cyberattacks executed in 2021 against Uranium Finance, a decentralized cryptocurrency exchange. Spalletta successfully exploited underlying flaws within the platform’s smart contracts to steal approximately $1.4 million in the initial intrusion, followed by a devastating $53.3 million secondary attack that ultimately forced the exchange to completely shut down operations.
Following the thefts, Spalletta reportedly laundered the stolen digital assets through a series of complex cryptocurrency transactions, routing funds through privacy-enhancing protocols such as Tornado Cash. Rather than liquidating the assets entirely into traditional fiat currency, a portion of the illicit proceeds was converted into high-value collectibles, including rare Magic: The Gathering and Pokémon cards, as well as antique Roman coins. Following the jury’s verdict, Spalletta faces a potential prison sentence of up to 10 years on the computer fraud count and up to 20 years on the money laundering count.
CISA Narrows Cyber Retention Pay Eligibility Criteria
On the domestic policy front, the Cybersecurity and Infrastructure Security Agency (CISA) has announced plans to maintain its Cybersecurity Retention Incentive program—which provides bonuses worth up to 25% of an employee’s base salary—through fiscal year 2027. However, the program is undergoing a significant administrative overhaul. Under the newly established criteria, staff members must hold a performance rating of "exceeds expectations" or higher and spend at least 51% of their working hours performing direct cyber duties within one of three designated job series.
For personnel operating outside those specific job series who nevertheless spend 75% or more of their time on specialized cyber work, an application process through a dedicated review board will be available. This policy shift follows a comprehensive audit by the Department of Homeland Security’s inspector general, which concluded that the previous iteration of the retention incentive program had been mismanaged and applied far too broadly across the agency.
Coalition Maps Out Proposed CISA Directive for Federal Operational Technology
Reflecting growing concerns over industrial cybersecurity, the Operational Technology Cybersecurity Coalition (OTCC) has released a formal proposal outlining what a CISA Binding Operational Directive (BOD) should explicitly require for operational technology (OT) across federal civilian agencies. Federal agencies currently rely on more than 8,000 General Services Administration-managed facilities that incorporate critical building management systems, including HVAC controls, power management grids, physical access control mechanisms, and advanced building automation networks.
To protect these physical environments from cyber intrusion, the OTCC proposal urges CISA to mandate that federal agencies designate a senior official or specialized office with direct accountability for OT security. Furthermore, the directive would require agencies to rigorously apply existing baseline security requirements and prioritize the implementation of CISA’s established Cybersecurity Performance Goals tailored specifically to industrial control systems.
Domino’s Resets Customer Accounts Hijacked via Credential Stuffing
Consumer-facing brands continue to grapple with automated credential abuse. Fast-food giant Domino’s has begun notifying a limited number of customers that their online accounts were accessed by unauthorized third parties. The intrusions were facilitated via credential stuffing attacks, where malicious actors automated login attempts using email address and password pairs that had been previously leaked in entirely unrelated third-party data breaches.
Domino’s emphasized that its internal corporate systems were not compromised during the incident, and noted that the platform does not store raw customer payment card details. Out of an abundance of caution, the company initiated forced password resets for all affected accounts to lock out unauthorized users and prevent further fraudulent activity.
AI Under Scrutiny as South Korea Probes Major Bank Cyberattacks
In international security news, South Korean President Lee Jae Myung publicly stated that investigators have found clear indicators suggesting artificial intelligence tools were leveraged in recent, highly coordinated cyberattacks targeting the nation’s banking sector. The incidents resulted in unauthorized access and subsequent breaches of sensitive customer personal information. Local law enforcement and national security authorities have launched a comprehensive investigation into the scope of the attacks.
While authorities have not yet publicly disclosed the specific AI models utilized or the complete financial and operational impact of the breaches, private security research has shed light on the infrastructure. Cybersecurity firm CrowdStrike reported finding artifacts such as Claude Code session histories, ARTEX configuration files, and Claude memory files while analyzing infrastructure tied to the attacks. Based on these technical findings, researchers hold moderate confidence that a Chinese-speaking, financially motivated threat actor group is driving the campaign.
Empire Market Co-Founder Sentenced to Four Decades in Federal Prison
The long-arm of federal law enforcement caught up with another major figure in the dark web ecosystem this week. Raheim Hamilton, a 30-year-old resident of Virginia, was sentenced to 40 years in federal prison and ordered to pay a $5 million financial penalty. Hamilton previously pleaded guilty to drug conspiracy charges relating to his role as a co-creator and administrator of Empire Market, a notorious dark web marketplace that operated extensively between 2018 and 2020 alongside co-founder Thomas Pavey.
During its period of operation, Empire Market facilitated more than four million individual transactions totaling over $430 million in illicit revenue. While the majority of sales centered on illegal narcotics, the marketplace also served as a clearinghouse for stolen credentials, compromised personal identifying information, counterfeit currency, and specialized hacking tools. Pavey, who entered a guilty plea in the case last year, is scheduled to receive his formal sentencing later this month.
Exposed Nvidia DCGM Exporters Leak Telemetry from Over 12,000 GPUs
Hardware and infrastructure monitoring tools frequently present unexpected attack surfaces when misconfigured. Security researchers disclosed details regarding CVE-2026-47483, a high-severity vulnerability impacting Nvidia’s Data Center GPU Manager (DCGM) Exporter tool, which is widely used to track graphics processing unit telemetry in high-performance computing environments.
The vulnerability allows unauthenticated, remote attackers to flood the tool’s internal profiling endpoints with specially crafted requests, triggering severe resource exhaustion and crashing the service. This disruption can severely degrade or halt critical artificial intelligence workloads running concurrently on the host system. During global scans conducted between March and May 2026, researchers discovered approximately 2,100 enterprise hosts inadvertently exposing the DCGM exporter directly to the public internet without authentication, thereby leaking granular performance and usage telemetry from more than 12,000 individual GPUs. Nvidia has since released security updates addressing the flaw, urging administrators to promptly upgrade to version 4.8.2 or later.
Shai-Hulud-Style Worm Slips into Tensorlake’s NPM SDK
Software package repository security faced another supply chain disruption after version 0.5.144 of the "tensorlake" package—the official npm SDK designed for Tensorlake’s AI agent sandboxes—was maliciously compromised. Security analysts at Socket and Sonatype classified the incident as a ChainDrop or Shai-Hulud style supply chain attack, noting that the compromised package was engineered to execute a credential-harvesting worm during the standard installation process.
The embedded malware aggressively scans infected developer machines to harvest authentication secrets across a wide array of development ecosystems, including npm, GitHub, AWS, Kubernetes, and HashiCorp Vault, alongside configuration files associated with modern AI coding assistants. Furthermore, the worm possesses the capability to execute arbitrary code provided by the external attacker and automatically republish modified malicious versions through other software packages managed by the victim, amplifying the downstream risk to the broader developer community.
