SecurityWeek’s weekly cybersecurity roundup provides a comprehensive look at critical developments across the global threat landscape. This week’s reports highlight a concerning trend of adversaries targeting non-traditional infrastructure, the emergence of systemic vulnerabilities in foundational server firmware, and the increasing sophistication of AI-powered offensive operations. As the digital and physical worlds continue to converge, these incidents underscore the fragility of interconnected systems and the urgent need for robust defensive postures.
Iran Suspected in US Military Freezer Hacks
US officials are investigating a series of unusual cyberattacks targeting the refrigeration systems at several US military base commissaries. According to reports from Nextgov, the malfunctions, which began in late August, resulted in the spoilage of food supplies at approximately six military facilities. Investigators are particularly focused on an incident at Fort Huachuca in Arizona, where a freezer was remotely manipulated into a defrost cycle, causing temperatures to rise and compromising the stored contents.
The investigation is a multi-agency effort, involving the US Cyber Command and Defense Department law enforcement. Notably, the incident mirrors a broader pattern of industrial control system (ICS) sabotage. Earlier this month, similar tactics were employed against approximately 350 commercial establishments in France, where attackers successfully overrode refrigeration controllers. The proximity of these incidents has led officials to suspect state-sponsored involvement, with intelligence pointing toward Iran-linked threat actors as the primary suspects.
Critical Firmware Flaw Exposes Global Server Infrastructure
A 16-year-old vulnerability, tracked as CVE-2026-97631, has emerged as a significant threat to global server infrastructure. The flaw exists within TianoCore EDK II, an open-source UEFI firmware framework that serves as the foundation for the majority of modern server vendors and data centers. By leveraging this code, major firmware suppliers—including AMI, Insyde, and Phoenix—have inadvertently distributed a systemic buffer overflow vulnerability that affects a vast percentage of the world’s servers.
Researchers from Anthropic and Google discovered the flaw, which allows an attacker with local network access to send a malicious, oversized ICMP error message to a machine during its network boot (PXE) process. Successful exploitation can hang the system or, if the specific firmware build is known, permit the execution of code at the highest privilege levels before security mechanisms like Secure Boot are initialized. While the flaw became public ahead of the planned January disclosure due to an early code fix published by AMI, organizations are now racing to secure their environments. The vulnerability highlights the inherent risks of shared codebases in critical infrastructure, emphasizing the difficulty of patching deep-layer firmware across fragmented global supply chains.
Anthropic and Cathedral Clash Over Security Research
A dispute has erupted regarding the scope and nature of a cybersecurity assessment conducted by Cathedral, a startup founded by former staff members of the Department of Government Efficiency (DOGE). A draft report reviewed by The Washington Post alleged that Cathedral’s AI agents successfully infiltrated Anthropic’s internal systems in under 12 hours. The startup claimed its agents bypassed defenses to access sensitive information, including meeting transcripts, salary data, and internal Slack discussions regarding the company’s most advanced artificial intelligence models.
Anthropic, which received these findings through its established bug bounty program, has pushed back against the startup’s characterization of the breach. The AI company maintains that the access was limited to a specific subset of sales, HR, and IT systems, and that its core intellectual property—including source code, model weights, and actual Claude conversation logs—remained secure. Furthermore, Anthropic terminated the startup’s access immediately upon discovery and has accused Cathedral of violating the terms of the bug bounty program by failing to provide full disclosure regarding the extent of their activity.
Election Infrastructure Stalls Following FEC Intrusion
The democratic process faced a logistical hurdle in the United States when the Federal Election Commission (FEC) was forced to take its electronic filing system offline. An internal email cited by Reuters indicated that an attempted breach necessitated the shutdown, which occurred just one week before the November 3 midterm elections. The outage prevented political organizations from submitting mandatory 24-hour and 48-hour reports, which are critical for transparency regarding last-minute campaign spending. While the FEC was able to restore the system by October 28, the incident highlights the vulnerability of administrative election infrastructure to even targeted disruption attempts.
Data Breaches at University of Copenhagen and ENGIE
Academic and corporate entities continue to struggle with data exposure. The University of Copenhagen reported that hackers gained unauthorized access to a test environment, resulting in the compromise of personal data belonging to approximately 525,000 individuals. The stolen information includes names, phone numbers, email addresses, and national identity (CPR) numbers. Additionally, PIN codes for identification cards used by 20,000 staff and students were exposed, forcing the university to conduct a mass reissuance of credentials.
In the energy sector, French giant ENGIE confirmed a data breach originating from a third-party tool used by its sales partners for prospecting. While ENGIE emphasized that its internal IT systems remained isolated and secure, the incident resulted in the theft of customer information, including contract identifiers and energy supply point details. Although no banking data was reported stolen, the breach underscores the risks associated with third-party software supply chains in the utility industry.
The Rise of AI-Powered Threats and Impersonation
The threat landscape is rapidly shifting as attackers integrate artificial intelligence into their toolsets. Cequence security researchers have identified a prolific threat actor deploying a network of over 300 IP addresses to scan the web for sensitive files like .env and SSH keys. This actor rotates through roughly 32 AI bot identities—including those impersonating legitimate services from OpenAI, Anthropic, Meta, and Perplexity—to evade detection. In some instances, this impersonation traffic accounted for 80% of all AI-labeled traffic observed by organizations.
Simultaneously, Cloudflare and the SGLang inference engine project have addressed a cross-tenant data leak. Between August 27 and September 9, 155 responses on Cloudflare’s Workers AI platform inadvertently contained data from other users’ previous requests. The leak, which occurred due to a configuration mismatch and an SGLang bug, affected prompts exceeding 681,280 tokens. Cloudflare has since remediated the issue and advised operators to update to SGLang version 0.5.22.
Sophisticated criminal groups are also accelerating their adoption of AI. The Gold Ogre group, responsible for the Aur0ra ransomware, is utilizing AI agents via Cursor to automate network discovery and lateral movement. Cisco Talos has also detailed "NIK C2," an AI-integrated framework capable of credential theft and ransomware deployment. Meanwhile, security researchers have observed "grey-market" relay services that reconfigure genuine CLI tools for platforms like Claude and Codex to route traffic through sandboxed-disabled, third-party servers.
Regulatory Compliance and Global Sanctions
Addressing the regulatory burden of the cybersecurity era, Rockwell Automation has expanded its SecureOT platform to assist OEMs in complying with the EU’s Cyber Resilience Act and Machinery Regulation. These tools provide machine-level inventory management and automated compliance reporting, helping manufacturers align with standards such as IEC 62443.
Finally, the US Department of Justice and the Treasury Department have intensified their efforts against North Korean IT worker schemes. John Allen Kurrle Jr., a defense contractor, pleaded guilty to wire fraud conspiracy for allowing North Korean nationals to use his identity to secure remote IT employment at US firms. In parallel, the Office of Foreign Assets Control (OFAC) sanctioned 10 individuals and one entity—the Russia-based First Financial and Technical Development Center—for their roles in facilitating the conversion of illicit crypto earnings and deploying North Korean workers to maintain a foothold in the Western technology sector.

