In Other News: AI Used in Korean Bank Breaches, Poem-Guided Botnet, Empire Admin Gets 40 Years
SecurityWeek roundup: PoeLLM botnet resolves C&C via GitHub poem, GhostAction hit 772 repos, CrowdStrike links Chinese actor to Korean bank breaches.
Black Lotus Labs detailed PoeLLM, active since at least April 2026, which mines cryptocurrency on exposed LiteLLM, Ollama, Gotenberg and Gitea services and derives its C&C IP from four keywords in a GitHub poem updated 11 times. GitGuardian says the GhostAction secret-stealing GitHub Actions workflow reached 772 repositories (373 users/orgs) and targeted 2,577 secrets between August 31 and September 30. CrowdStrike assessed with moderate confidence that a Chinese-speaking financially motivated actor is behind South Korean bank breaches where AI use is suspected, and researchers disclosed CVE-2026-47483 in Nvidia's DCGM Exporter, with about 2,100 unauthenticated hosts leaking telemetry from more than 12,000 GPUs (fixed in 4.8.2). Additional items include the compromised tensorlake npm SDK 0.5.144 credential-stealing worm and Empire Market co-founder Raheim Hamilton's 40-year sentence.
- PoeLLM botnet mines crypto on exposed LiteLLM, Ollama, Gotenberg, Gitea; C&C hidden in GitHub poem.
- GhostAction secret-stealing workflow spread to 772 repos, targeting 2,577 secrets.
- CrowdStrike: Chinese-speaking financially motivated actor likely behind Korean bank breaches; AI suspected.
- CVE-2026-47483 in Nvidia DCGM Exporter; ~2,100 exposed hosts leaked telemetry from 12,000 GPUs.
- tensorlake npm SDK 0.5.144 compromised with credential-stealing Shai-Hulud worm.
Vulnerabilities mentionedAll →
- CVE-2026-474838.2<1%NVIDIA DCGM Exporter for all platforms contains a vulnerability in the /debug/pprof endpoints, where an attacker could cause uncontrolled resource consumption…published
| CVE | Vulnerability | CVSS | EPSS | Flags | Affected | Exposure | Published |
|---|---|---|---|---|---|---|---|
| CVE-2026-47483 | NVIDIA DCGM Exporter for all platforms contains a vulnerability in the /debug/pprof endpoints, where an attacker could cause uncontrolled resource consumption… NVIDIA DCGM Exporter for all platforms contains a vulnerability in the /debug/pprof endpoints, where an attacker could cause uncontrolled resource consumption by submitting concurrent unauthenticated profiling requests. A successful exploit of this vulnerability might lead to denial of service and information disclosure. |
Full article972 words · extracted from securityweek.com · click to collapse
SecurityWeek’s weekly cybersecurity news roundup offers a concise overview of important developments that may not receive full standalone coverage yet remain relevant to the broader threat landscape.
This curated summary highlights key stories across vulnerability disclosures, emerging attack methods, policy updates, industry reports, and other noteworthy events to help readers stay well-informed about the evolving cybersecurity environment.
Here are this week’s highlights:
PoeLLM malware finds its C&C server hidden in a GitHub poem
Black Lotus Labs has detailed PoeLLM, a malware active since at least April 2026 that targets exposed AI and open-source services (mainly LiteLLM, Ollama, Gotenberg and Gitea) to mine cryptocurrency and expand its botnet. Infected machines extract four keywords from a poem hosted on GitHub and convert them into the IP address of the current C&C server, so the operator can switch servers by changing those words. The operator, assessed to be Italian-speaking, has updated the poem 11 times.
Advertisement. Scroll to continue reading.
GhostAction secret theft spreads to 772 more GitHub repos
GitGuardian says the GhostAction supply chain campaign pushed its secret-stealing GitHub Actions workflow to 772 public repositories (belonging to 373 users and organizations) between August 31 and September 30, targeting 2,577 secrets such as SSH keys and Azure, AWS and database credentials. Apart from a new exfiltration server, the attacker reused the 2025 playbook, and GitGuardian’s data shows the campaign never really stopped.
Jury convicts Uranium Finance hacker
A jury has convicted Jonathan Spalletta, 36, of Maryland, of computer fraud and money laundering over two 2021 hacks of decentralized crypto exchange Uranium Finance. He abused smart contract flaws to take roughly $1.4 million and then $53.3 million (the second attack forced Uranium to shut down), laundered the funds through a series of crypto transactions that included Tornado Cash, and used them to buy collectibles such as rare Magic: The Gathering and Pokémon cards and antique Roman coins. He faces up to 10 years in prison on the fraud count and 20 years on the money laundering count.
CISA narrows cyber retention pay
CISA will keep its Cybersecurity Retention Incentive program (worth up to 25% of base salary) through fiscal 2027, but under new criteria staff must hold an “exceeds expectations” or higher rating and spend at least 51% of their time on cyber duties in one of three job series. Those outside the series who spend 75% or more on cyber work can apply to a review board. The overhaul follows a DHS inspector general finding that the program was mismanaged and applied too broadly.
Coalition maps out what a CISA directive for federal OT should require
The Operational Technology Cybersecurity Coalition (OTCC) has published a proposal for a CISA BOD focused solely on OT at federal civilian agencies, which rely on more than 8,000 GSA-managed facilities with HVAC, power management, access control and building automation systems. The group wants the directive to require agencies to designate a senior official or office accountable for OT security, apply relevant existing requirements, and prioritize CISA’s Cybersecurity Performance Goals.
Domino’s resets accounts hijacked with recycled passwords
Domino’s is telling a small number of customers that their accounts were accessed by an unauthorized third party through credential stuffing, using email and password pairs leaked in unrelated breaches. The company says its systems were not compromised and it doesn’t store payment details, but it has reset the affected accounts.
AI under suspicion as South Korea probes bank cyberattacks
South Korean President Lee Jae Myung said there are signs that AI was used in some of the recent hacking incidents targeting the country’s banks. Police have reportedly launched a full-scale investigation into the attacks, which led to a breach of customers’ personal information. Authorities have yet to reveal which AI tools were used or the full scale of the breaches. CrowdStrike this week reported finding Claude Code session histories, ARTEX configuration files, and Claude memory files while analyzing the attack infrastructure. The security firm believes with moderate confidence that a Chinese-speaking financially motivated threat actor is likely behind the attacks.
Empire Market co-founder gets four decades behind bars
Raheim Hamilton, 30, of Virginia, has been sentenced to 40 years in prison and fined $5 million after pleading guilty to a drug conspiracy charge over Empire Market, the dark web marketplace he co-created and ran with Thomas Pavey from 2018 to 2020. The site handled more than four million transactions worth over $430 million, mostly drug sales, and also sold stolen credentials and personal information, counterfeit currency and hacking tools. Pavey, who pleaded guilty last year, is scheduled to be sentenced later this month.
Exposed Nvidia DCGM exporters leak telemetry from 12,000 GPUs
Researchers have disclosed CVE-2026-47483, a high-severity flaw in Nvidia’s DCGM Exporter GPU monitoring tool. Unauthenticated attackers can flood its profiling endpoints with requests to exhaust resources and crash the service, potentially slowing AI workloads running on the same host. In scans between March and May 2026, the researchers found roughly 2,100 hosts exposing the exporter to the internet without authentication, leaking telemetry from more than 12,000 GPUs. Nvidia has addressed the flaw, and users are advised to update to version 4.8.2 or later.
Shai-Hulud-style worm slips into Tensorlake’s npm SDK
Version 0.5.144 of tensorlake, the npm SDK for Tensorlake’s AI agent sandboxes, was compromised to run a credential-stealing worm during installation, in what Socket describes as a ChainDrop/Shai-Hulud supply chain attack. Socket and Sonatype say the malware harvests npm, GitHub, AWS, Kubernetes and Vault credentials, as well as AI coding tool configurations. It can execute code supplied by the attacker and republish itself through other packages the victim can publish.
Related: In Other News: Clop Leak Site Takeover, Docker Botnet Hunts AI Keys, Water Utility Exposure
Related: In Other News: $15K iCloud Spoofing Bugs, AI Policy Experts Phished, Adblocker Spies on AI Chats