# Weekly Threat Roundup: AI Banking Attacks & Nvidia Vulnerability

> Explore this week's key cybersecurity news, including PoeLLM botnets, South Korean banking attacks, and Nvidia GPU monitoring flaws.

- Published: 2026-10-09T14:50:02.000Z
- Severity: high
- Category: Threat Intel
- Tags: PoeLLM, NVIDIA, Supply Chain Attack, Credential Theft, Ransomware
- CVEs: CVE-2026-47483 (CVSS 7.5)
- Author: Runtime Rebel Intel
- Primary source: https://www.securityweek.com/in-other-news-ai-used-in-korean-bank-breaches-poem-guided-botnet-empire-admin-gets-40-years/
- Canonical: https://runtimerebel.com/blog/weekly-threat-roundup-ai-banking-attacks-nvidia-vulnerability

## Key points

- Immediate impact: organizations face emerging threats across software supply chains, exposed AI infrastructure, and active credential-stuffing campaigns.
- Affected systems: exposed AI management tools, GitHub repositories, and Nvidia DCGM Exporter versions prior to 4.8.2.
- Remediation: update vulnerable exporter instances immediately and audit public repositories for unauthorized workflow modifications.

## Weekly Cybersecurity Intelligence Roundup

Security professionals must navigate an increasingly diverse [threat landscape](/glossary#threat-landscape) spanning automated botnets, sophisticated supply chain compromises, and hardware-level telemetry exposures. According to [SecurityWeek](https://www.securityweek.com/in-other-news-ai-used-in-korean-bank-breaches-poem-guided-botnet-empire-admin-gets-40-years/), recent intelligence highlights several major developments requiring immediate attention from defenders and security architects alike.

### Emerging [Malware](/glossary#malware) and Supply Chain Vectors

Among the newly detailed threats is **PoeLLM**, a piece of malware active since April 2026 targeting exposed artificial intelligence and open-source services such as LiteLLM, Ollama, Gotenberg, and Gitea. The malware uses an innovative command-and-control mechanism by extracting four keywords from a poem hosted on GitHub to dynamically resolve its C&C server IP address. Operators have modified the reference poem eleven times to reroute traffic.

Concurrently, supply chain vectors continue to plague development environments. The **GhostAction** campaign pushed secret-stealing workflows to 772 public GitHub repositories, targeting thousands of credentials including [SSH](/glossary#secure-shell-ssh) keys, AWS tokens, and database secrets. Additionally, Tensorlake's npm SDK version 0.5.144 was compromised to run a credential-stealing [worm](/glossary#worm) during installation, echoing previous automated software supply chain propagation methods.

### Infrastructure Vulnerabilities and Enforcement

Hardware telemetry and monitoring tools have also come under scrutiny. Researchers disclosed [CVE-2026-47483](https://nvd.nist.gov/vuln/detail/CVE-2026-47483), a high-severity denial-of-service [vulnerability](/glossary#vulnerability) affecting Nvidia's DCGM Exporter GPU monitoring tool. Unauthenticated attackers can flood profiling endpoints to exhaust system resources and crash the service, disrupting high-performance [AI](/glossary#ai) workloads. Internet-wide scans revealed over 2,100 exposed hosts leaking telemetry from more than 12,000 GPUs.

In policy and law enforcement updates, South Korean authorities have launched investigations into bank cyberattacks, with preliminary findings indicating potential use of artificial intelligence in orchestrating the intrusions. Meanwhile, [dark web](/glossary#dark-web) infrastructure suffered another blow as a federal jury convicted a marketplace co-founder, resulting in a 40-year prison sentence.

### Actionable Recommendations for Security Teams

Defenders should prioritize the following mitigation steps to protect enterprise environments:

* **[Patch](/glossary#patch) Monitoring Tools:** Upgrade Nvidia DCGM Exporter to version 4.8.2 or later to address [CVE-2026-47483](https://nvd.nist.gov/vuln/detail/CVE-2026-47483) and restrict network access to telemetry endpoints.
* **Audit Developer Ecosystems:** Review public and private code repositories for unauthorized GitHub Actions workflows or malicious modifications in package manager dependencies.
* **Secure AI Infrastructure:** Ensure that internal instances of [LLM](/glossary#jailbreak-llm) runners, Gitea, and related utilities are not exposed directly to the public internet without proper authentication layers.

**Related:** [Emerging Cyber Threats and Espionage Risks in Neurotechnology](/blog/emerging-cyber-threats-and-espionage-risks-in-neurotechnology), [GTIG AI Threat Tracker: Evolution of Adversarial Agentic AI](/blog/gtig-ai-threat-tracker-evolution-of-adversarial-agentic-ai)

---

AI-generated analysis from the primary source above; not human-reviewed before publication — verify anything operational against the original (https://runtimerebel.com/editorial). Quote with attribution and a link to the canonical URL: https://runtimerebel.com/blog/weekly-threat-roundup-ai-banking-attacks-nvidia-vulnerability