In Other News: AI Used in Korean Bank Breaches, Poem-Guided Botnet, Empire Admin Gets 40 Years

A cybersecurity roundup highlights several concurrent developments in the threat landscape, ranging from specific malware operations and legal outcomes to policy shifts and emerging AI-related risks. A critical development involves a PoeLLM malware that hides its Command and Control (C&C) server within GitHub poems, allowing operators to change servers by modifying keywords in a poem hosted on GitHub. Simultaneously, supply chain exposure is noted through the GhostAction campaign, which distributed secret-stealing GitHub Actions workflows across numerous repositories, resulting in the exfiltration of sensitive credentials from 772 public repositories.
Legal and financial enforcement has also been demonstrated, as a jury convicted a hacker for exploiting smart contract flaws in decentralized crypto exchanges to steal significant funds and subsequently launder them through various cryptocurrency transactions. In areas concerning federal cybersecurity governance, CISA is adjusting its Cybersecurity Retention Incentive program criteria, requiring staff to meet specific performance benchmarks and spend time on cyber duties to maintain eligibility. Furthermore, discussions around operational technology (OT) security are evolving, with the Operational Technology Cybersecurity Coalition proposing a CISA directive focused specifically on securing facilities managing building automation and power systems in federal agencies.
Finally, risks related to AI and supply chain integrity surface through investigations into bank cyberattacks where AI tools may have been involved, and a discovery concerning an Nvidia monitoring tool leak exposing telemetry from numerous GPUs. These events collectively underscore ongoing challenges in securing software supply chains, managing advanced persistent threats against digital assets, and establishing cohesive governance frameworks for critical infrastructure environments amid rapid technological evolution.

* PoeLLM malware hides its C&C server in a GitHub poem.
* Black Lotus Labs detailed PoeLLM, which targets exposed AI/open-source services (LiteLLM, Ollama, Gotenberg, Gitea) for cryptocurrency mining and botnet expansion.
* Infected machines extract four keywords from a GitHub poem to determine the current C&C server IP address.
* GhostAction supply chain campaign pushed secret-stealing GitHub Actions workflows to 772 public repositories between August 31 and September 30, targeting 2,577 secrets (e.g., SSH keys, Azure/AWS credentials).
* A jury convicted Jonathan Spalletta of computer fraud and money laundering over two hacks of Uranium Finance, involving smart contract flaws and fund laundering via Tornado Cash.
* CISA will maintain the Cybersecurity Retention Incentive program through fiscal 2027, contingent on staff achieving "exceeds expectations" ratings and dedicating at least 51% of time to cyber duties in specific job series.
* The Operational Technology Cybersecurity Coalition proposed a CISA directive focused solely on OT security for federal civilian agencies.
* Domino’s notified customers that accounts were accessed via credential stuffing using leaked email/password pairs from unrelated breaches, and accounts were reset.
* South Korean President Lee Jae Myung indicated signs that AI was used in recent bank hacking incidents; CrowdStrike found Claude Code session histories during analysis.
* Raheim Hamilton was sentenced to 40 years in prison and fined $5 million for a drug conspiracy related to the Empire Market dark web marketplace.
* Researchers disclosed CVE-2026-47483 in Nvidia’s DCGM Exporter GPU monitoring tool, allowing unauthenticated attackers to exhaust resources by flooding endpoints, leaking telemetry from over 12,000 GPUs.
* A credential-stealing worm was found in tensorlake's npm SDK version 0.5.144 during installation, harvesting credentials from various services.

The convergence of narratives reveals a systemic tension between the rapid evolution of open-source and AI technologies, the persistence of traditional criminal exploitation methods adapted for new vectors, and the struggle to establish coherent regulatory and governance structures. The PoeLLM example illustrates how seemingly innocuous creative outputs (poetry) can be weaponized within developer ecosystems to establish resilient C&C infrastructure, pointing toward a challenge in securing the context of code and content, not just the code itself. This links directly to the supply chain attacks observed with GhostAction and the Tensorlake worm, suggesting that integrity is broken at the foundational layers—from dependency management (npm) to the deployment environment (GitHub Actions).
The legal case involving Uranium Finance highlights the monetization potential derived from exploiting systemic flaws in decentralized financial infrastructure, demonstrating how high-value assets can be abstracted and laundered through complex, layered transactions. This mirrors the broader theme that technological complexity often serves as a shield for sophisticated financial crime. The speculation surrounding AI use in South Korean bank attacks suggests an emerging vector where the opacity of machine learning models might become an exploitable blind spot, shifting focus from explicit code vulnerabilities to inherent model trustworthiness.
The policy and infrastructure discussions—CISA adjustments and the OTCC proposal—indicate an attempt to layer traditional security mandates onto newly recognized risk domains (OT) and evolving operational realities. The pattern emerging is one of reactive adaptation: sophisticated threat actors leverage bleeding-edge technology (AI, open source) for stealthy operations while governing bodies attempt to impose structured accountability on established systems. The implication for agency is that resilience requires not just patching vulnerabilities but establishing epistemological control over the tools and data used in development and finance, questioning who controls the context surrounding these innovations. What mechanisms are needed to audit the provenance of creative outputs and AI-assisted code generation before they become exploitable attack surfaces? Where does accountability reside when exploits traverse layers from a GitHub poem to a financial exchange?