In other news: AI used in Korean bank breaches, Poem-driven botnet, Empire administrator gets 40 years in prison

Weekly from SecurityWeek Cybersecurity News Summary provides a succinct overview of key developments that may not be covered entirely in isolation but are nonetheless relevant to the broader threat landscape.

This curated summary highlights key stories about vulnerability disclosures, new attack vectors, policy updates, industry reports and other notable 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 described in detail PoeLLMa malware active since at least April 2026 that targets exposed AI and open source services (mainly LiteLLM, Ollama, Gotenberg and Gitea) to mine cryptocurrencies 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, allowing the operator to change servers by changing these words. The operator, classified as Italian-speaking, has updated the poem eleven times.

Advertising. Scroll to continue reading.

GhostAction’s secret theft spreads to 772 other GitHub repos

GitGuardian says so GhostAction The Supply Chain campaign expanded its GitHub Actions secret-stealing workflow to 772 public repositories (from 373 users and organizations) between August 31 and September 30, targeting 2,577 secrets such as SSH keys and Azure, AWS, and database credentials. Aside from a new exfiltration server, the attacker reused the 2025 playbook, and GitGuardian data shows the campaign was never actually stopped.

Jury convicts Uranium Finance hacker

A jury has convicted Jonathan Spalletta, 36, of Maryland, of computer fraud and money laundering in two 2021 hacks of the decentralized crypto exchange Uranium Finance. He exploited smart contract vulnerabilities to collect about $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, including Tornado Cash, and used them to purchase collectibles such as rare Magic: The Gathering and Pokémon cards and ancient Roman coins. He faces Up to 10 years in prison for fraud and 20 years for money laundering.

CISA limits cyber retention compensation

CISA is retained its Cybersecurity Retention Incentive program (worth up to 25% of base salary) through fiscal year 2027, but new criteria requires employees to have an “exceeds expectations” rating or higher and spend at least 51% of their time on cyber tasks in one of three job lines. Those outside the series who spend 75% or more on cyber work can apply to a review board. The revision follows a DHS inspector general’s finding that the program was mismanaged and overly broad.

The coalition sets out what a federal CISA policy for OT should require

The Operational Technology Cybersecurity Coalition (OTCC) has published one Suggestion for a CISA BOD focused exclusively on OT at federal civilian agencies that rely on more than 8,000 GSA-managed facilities with HVAC, energy management, access control and building automation systems. The group wants the policy to require agencies to designate a senior official or entity responsible for OT security, apply relevant existing requirements, and prioritize CISA cybersecurity performance goals.

Domino’s resets accounts hijacked with recycled passwords

Domino’s is telling a small number of customers that their accounts have been accessed by unauthorized third parties Credential stuffingusing email and password pairs leaked in unrelated security breaches. The company says its systems have not been compromised and that it does not store payment details, but has reset the affected accounts.

AI under suspicion as South Korea investigates cyberattacks on banks

South Korean President Lee Jae Myung said there were signs that AI had been used in some of the recent cases Hacking incidents The aim is to target the country’s banks. Police have reportedly launched a full investigation into the attacks, which resulted in a breach of customers’ personal information. Authorities have not yet disclosed which AI tools were used or the scale of the breaches. CrowdStrike this week reported Finding Claude code session histories, ARTEX configuration files, and Claude save files when analyzing attack infrastructure. The security company says with moderate confidence that a Chinese-speaking, financially motivated threat actor is likely behind the attacks.

Empire Market co-founder has been behind bars for four decades

Raheim Hamilton, 30, of Virginia, was sentenced was sentenced to 40 years in prison and a $5 million fine after pleading guilty to a drug conspiracy related to Empire Market, the dark web marketplace he co-founded and operated with Thomas Pavey from 2018 to 2020. The site processed 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 data from 12,000 GPUs

researchers have disclosed CVE-2026-47483a fatal flaw in Nvidia’s GPU monitoring tool DCGM Exporter. Unauthenticated attackers can flood the profiling endpoints with requests to exhaust resources and crash the service, potentially slowing down AI workloads running on the same host. In scans between March and May 2026, researchers found approximately 2,100 hosts that exposed the exporter to the Internet without authentication, leaking telemetry data from more than 12,000 GPUs. Nvidia has fixed the bug 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 execute a credential stealing worm during installation, in what Socket describes as a ChainDrop/Shai-Hulud supply chain attack. socket And Sonatype Suppose the malware collects NPM, GitHub, AWS, Kubernetes, and Vault credentials, as well as AI coding tool configurations. It can execute code provided by the attacker and republish itself via other packages that the victim can publish.

Related: In other news: Clop leak site takeover, Docker botnet hunts AI keys, water utility compromise

Related: In other news: $15,000 iCloud spoofing bugs, AI policy experts phishing, Adblocker spies on AI chats

Leave a Comment