AI Under Investigation in South Korean Bank Cyberattacks
South Korean President Lee Jae Myung stated there are indications that artificial intelligence was used in some of the recent hacking incidents targeting the country’s banks. Authorities have launched a full-scale investigation into these attacks, which resulted in the breach of customers’ personal information. While the specific AI tools employed and the full scale of the data exposure remain undisclosed, CrowdStrike reported finding evidence during infrastructure analysis, including Claude Code session histories, ARTEX configuration files, and Claude memory files. The security firm assesses with moderate confidence that a Chinese-speaking, financially motivated threat actor is likely responsible for the intrusions.
PoeLLM Malware Uses GitHub Poem for C&C Communication
Black Lotus Labs revealed details about PoeLLM, a malware strain active since at least April 2026 that targets exposed AI and open-source services such as LiteLLM, Ollama, Gotenberg, and Gitea. The malware infects systems to mine cryptocurrency and expand its botnet. A notable feature of PoeLLM is its use of a poem hosted on GitHub as a covert command-and-control (C&C) mechanism. Infected machines extract four specific keywords from the poem and convert them into the IP address of the current C&C server, allowing the operator to redirect traffic simply by altering those words. The operator, believed to be Italian-speaking, has updated the poem 11 times to maintain control.
GhostAction Campaign Expands to 772 GitHub Repositories
GitGuardian reported that the GhostAction supply chain campaign successfully pushed its secret-stealing GitHub Actions workflow to 772 additional public repositories between August 31 and September 30, 2026. These repositories belong to 373 users and organizations and were targeted to exfiltrate 2,577 sensitive secrets, including SSH keys, AWS and Azure credentials, and database passwords. Although the attackers introduced a new exfiltration server, they largely reused tactics from their 2025 playbook. GitGuardian’s data indicates the campaign remained active throughout the period without significant interruption.
Jury Convicts Uranium Finance Hacker in $54.7 Million Crypto Theft
A federal jury convicted Jonathan Spalletta, 36, of Maryland, on charges of computer fraud and money laundering related to two separate hacks of the decentralized exchange Uranium Finance in 2021. Spalletta exploited smart contract vulnerabilities to first steal approximately $1.4 million, then later $53.3 million in a second attack that forced Uranium Finance to cease operations. He laundered the proceeds through a series of cryptocurrency transactions, including the use of Tornado Cash, and used the funds to purchase collectibles such as rare Magic: The Gathering and Pokémon cards, as well as antique Roman coins. Spalletta faces a maximum penalty of 10 years for fraud and 20 years for money laundering.
CISA Adjusts Cyber Retention Incentive Program Through 2027
The Cybersecurity and Infrastructure Security Agency (CISA) announced it will continue its Cybersecurity Retention Incentive program through fiscal year 2027, offering eligible employees up to 25% of their base salary. However, access to the benefit is now restricted: staff must hold a performance rating of "exceeds expectations" or higher and dedicate at least 51% of their work time to cybersecurity duties within one of three designated job series. Employees outside these series who spend 75% or more of their time on cyber-related tasks may petition a review board for eligibility. The revision follows a Department of Homeland Security inspector general report that found the original program was mismanaged and applied too broadly across the workforce.
OTCC Proposes Federal OT Security Directive for CISA
The Operational Technology Cybersecurity Coalition (OTCC) has submitted a proposal to CISA for a Binding Operational Directive (BOD) focused exclusively on operational technology (OT) security across federal civilian agencies. The coalition notes that these agencies manage over 8,000 facilities via the General Services Administration (GSA), encompassing systems such as HVAC, power distribution, access control, and building automation. The proposed directive would require agencies to appoint a senior official or office accountable for OT security, enforce compliance with existing relevant standards, and prioritize alignment with CISA’s Cybersecurity Performance Goals to strengthen resilience against cyber threats.
Domino’s Resets Accounts After Credential Stuffing Incidents
Domino’s Pizza notified a limited number of customers that their accounts had been accessed by unauthorized parties through credential stuffing attacks, where attackers used email and password combinations leaked in unrelated data breaches. The company emphasized that its own systems were not compromised and that it does not store payment card information. In response, Domino’s reset the affected accounts to prevent further unauthorized access and advised users to adopt unique, strong passwords.
Empire Market Co-Founder Sentenced to 40 Years in Prison
Raheim Hamilton, 30, of Virginia, was sentenced to 40 years in prison and fined $5 million after pleading guilty to a drug conspiracy charge tied to his role as co-founder and operator of Empire Market, a dark web marketplace active from 2018 to 2020. Hamilton ran the platform alongside Thomas Pavey, during which time it facilitated over four million transactions valued at more than $430 million. The majority of sales involved illicit drugs, but the marketplace also distributed stolen credentials, personal data, counterfeit currency, and hacking tools. Pavey, who previously pleaded guilty, is awaiting sentencing later in the month.
CVE-2026-47483: Unauthenticated Nvidia DCGM Exposer Exposes 12,000 GPUs
Researchers disclosed CVE-2026-47483, a high-severity vulnerability in Nvidia’s Data Center GPU Manager (DCGM) Exporter tool. The flaw allows unauthenticated attackers to send excessive requests to the exporter’s profiling endpoints, leading to resource exhaustion and potential service disruption. This could impair AI workloads running on the same host. Between March and May 2026, scans revealed approximately 2,100 internet-exposed hosts running the vulnerable exporter without authentication, thereby leaking telemetry from over 12,000 GPUs. Nvidia has since patched the issue and recommends users upgrade to version 4.8.2 or later to mitigate risk.
Shai-Hulud-Style Worm Found in Compromised Tensorlake npm SDK
Version 0.5.144 of the tensorlake npm package — the software development kit for Tensorlake’s AI agent sandbox environment — was found to contain a malicious worm during installation, characterized by Socket as a ChainDrop/Shai-Hulud-style supply chain attack. Upon execution, the malware harvests sensitive credentials from npm, GitHub, AWS, Kubernetes, and HashiCorp Vault, in addition to collecting configurations from AI coding tools. It can also execute arbitrary code supplied by the attacker and republish itself through downstream packages that victims may later distribute, enabling further propagation.
Source: SecurityWeek