SecurityWeek’s weekly cybersecurity roundup provides a comprehensive look at the critical developments shaping the global threat landscape. As cyberattacks grow in both sophistication and scope, this summary synthesizes essential news across vulnerability disclosures, state-sponsored campaigns, and the emerging intersection of artificial intelligence and malicious activity.
Suspected Iranian Involvement in US Military Freezer Sabotage
In a concerning development for critical infrastructure security, US officials are currently investigating a series of cyber-incidents involving military base commissaries. Reports indicate that freezer malfunctions, which led to the spoilage of food supplies, began occurring in late August across approximately six US military installations. One specific case at Fort Huachuca in Arizona saw a refrigeration unit forcibly switched into defrost mode, causing the temperature to rise and destroying the stored goods.
The investigation is being spearheaded by Cyber Command and Defense Department law enforcement units, with assistance from French authorities. The international cooperation stems from a strikingly similar pattern of attacks reported earlier this month, where hackers managed to override refrigeration controls at roughly 350 establishments across France. While the investigation is ongoing, authorities suspect an Iran-linked actor, highlighting the increasing vulnerability of internet-connected industrial and commercial hardware.
Critical Firmware Vulnerability Impacts Global Server Infrastructure
A 16-year-old buffer overflow vulnerability, identified as CVE-2026-97631, has sent shockwaves through the data center industry. The flaw resides within TianoCore EDK II, an open-source UEFI firmware framework that serves as the foundation for the majority of the world’s server vendors. Because this code is utilized by major firmware suppliers—including AMI, Insyde, and Phoenix—the impact of the vulnerability is near-universal.
Discovered by researchers at Anthropic and Google, the flaw allows an attacker with local network access to send a specially crafted, oversized ICMP error message to a server while it is performing a network boot (PXE). If the attacker is familiar with the target’s specific firmware build, they can cause the system to hang or, more critically, execute arbitrary code at the highest privilege level before security features like Secure Boot are initialized. While only systems utilizing network boot are currently at risk, the public disclosure occurred months ahead of the intended January date after AMI released a patch. Organizations are now racing to secure their infrastructure by applying firmware updates provided by their respective server manufacturers.
Dispute Over Anthropic’s Security Breach
The cybersecurity community is closely watching a dispute between the AI firm Anthropic and Cathedral, a startup founded by former DOGE staffers. A draft report, reviewed by the Washington Post, alleges that Cathedral’s AI agents successfully infiltrated Anthropic’s internal systems in under 12 hours. The researchers claim to have accessed sensitive data, including meeting transcripts, salary information, job offers, and internal discussions regarding the company’s cutting-edge AI models.
Anthropic has publicly contested the scope of the incident. While confirming that Cathedral accessed a limited number of HR, sales, and IT systems via third-party business software vulnerabilities, the company stated that it terminated the startup’s access immediately upon discovery. Anthropic maintains that its core assets—including source code, model weights, and actual Claude conversations—remained secure. Furthermore, the company has accused Cathedral of violating the terms of its bug bounty program by failing to fully disclose their activities, sparking a debate over the ethics of security research in the age of autonomous AI agents.
Election Security and FEC System Outage
The security of democratic processes faced a direct challenge as the US Federal Election Commission (FEC) was forced to take its electronic filing system offline following an attempted breach. Internal communications revealed that the intrusion attempt left political organizations unable to submit crucial campaign finance reports just one week before the November 3 midterm elections.
The outage specifically hindered the submission of 24-hour and 48-hour reports, which are designed to provide transparency regarding last-minute, high-volume campaign spending. While the FEC ensured that historical filings remained accessible to the public, the disruption caused significant concern regarding the integrity of campaign reporting during the critical final days of the election cycle. The agency successfully restored the system on October 28, following the incident.
Data Breaches at University of Copenhagen and ENGIE
Institutional and corporate data security remains under pressure following two significant breaches. The University of Copenhagen reported that hackers gained access to a test environment, compromising the personal data of approximately 525,000 individuals, including students and staff. Stolen information included names, phone numbers, email addresses, and national identity (CPR) numbers. Additionally, the PIN codes for roughly 20,000 ID cards were compromised, necessitating an immediate reissue of credentials. The university has not yet identified the perpetrators or the motivation behind the attack.
Simultaneously, French energy giant ENGIE disclosed a breach involving a tool used by its sales partners for customer prospecting. While the company confirmed that its internal IT systems were not compromised, the attackers successfully exfiltrated data belonging to customers in France. The stolen information includes contact details, customer IDs, and contract numbers, though no banking data was involved. ENGIE continues to investigate the scope of the incident as it works to notify affected individuals.
The Weaponization of AI in Cyberattacks
The threat landscape is undergoing a transformation as malicious actors increasingly leverage AI to automate and scale their operations. Cequence researchers have identified a persistent threat actor rotating through 32 different AI bot identities—impersonating tools from OpenAI, Meta, and others—to scan for exposed sensitive files like .env configurations and SSH keys. In some cases, these impersonators accounted for over 80% of an organization’s reported AI bot traffic.
Other firms are observing even more direct applications of AI in malware development. Sophos has detailed the Gold Ogre group, which uses AI-orchestrated agents to handle network discovery and lateral movement for its Aur0ra ransomware. Meanwhile, Cisco Talos has identified "NIK C2," an AI-integrated command-and-control framework capable of stealing credentials and deploying ransomware. Additionally, security firms have noted the rise of "grey-market" relay services that reconfigure genuine AI coding assistants to route requests through unauthorized servers, bypassing essential safety sandboxing.
Cloudflare Addresses AI Inference Leak
In the realm of cloud and AI infrastructure, Cloudflare recently addressed a cross-tenant data leak involving its Workers AI platform. Between August 27 and September 9, 155 responses generated by the GLM 5.3 model contained data from other customers’ previous requests. The leak was caused by a configuration mismatch in the model’s KV cache combined with a bug in the SGLang inference engine, which allowed requests to inadvertently access cached data from earlier sessions. Cloudflare has since implemented a fix and advises all SGLang operators to upgrade to version 0.5.22 or later.
North Korean IT Worker Scheme Prosecutions
The US Department of Justice continues to crack down on elaborate schemes designed to fund the North Korean regime. John Allen Kurrle Jr., a defense contractor employee with a Top Secret clearance, has pleaded guilty to wire fraud conspiracy for his role in an IT worker scheme. Kurrle allowed North Korean nationals to use his identity to secure remote jobs with US firms and hosted the victims’ company laptops at his residence.
This legal action coincides with new sanctions from the Office of Foreign Assets Control (OFAC) against 10 individuals and one entity. These sanctions target the financial infrastructure supporting North Korean IT workers, including Russian nationals accused of converting crypto-earnings into fiat currency and a Russia-based technical center that deploys these workers internationally.
Rockwell Automation Bolsters Industrial Security
As industrial manufacturers face mounting regulatory pressure, Rockwell Automation has expanded its "SecureOT" offering. The new capabilities are specifically designed to help original equipment manufacturers (OEMs) comply with the European Union’s Cyber Resilience Act (CRA) and updated Machinery Regulations. By providing machine-level hardware and firmware inventories, vulnerability tracking, and compliance reporting, Rockwell aims to help manufacturers align with international standards like IEC 62443, ensuring that security is integrated into the industrial supply chain from the outset.






