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:
Iran suspected in US military freezer hacks
US officials suspect Iran-linked hackers caused freezer malfunctions that spoiled food at around six military base commissaries starting in late August, Nextgov reports, citing people familiar with the matter. At Fort Huachuca in Arizona, a freezer entered defrost mode and heated its contents, and Cyber Command and Defense Department law enforcement units are investigating with French authorities, after hackers overrode refrigeration controls at some 350 French establishments earlier this month.
Most servers rely on UEFI code hit by critical boot flaw
A critical 16-year-old buffer overflow (CVE-2026-97631) in TianoCore EDK II, open source code that most of the world’s server vendors and data centers use through firmware suppliers such as AMI, Insyde and Phoenix, means the flaw affects most servers worldwide, Ars Technica reports. An attacker with local network access can send an oversized ICMP error message to a machine while it boots over the network (PXE) to hang the process or, if the target’s firmware build is known, run code at the highest privilege level before defenses such as Secure Boot take effect. Only machines that use network boot are exposed, and the flaw, found by Anthropic and Google researchers, became public months ahead of a planned January disclosure after AMI published a code fix, leaving organizations to seek firmware updates from their vendors.
Anthropic disputes scope of ex-DOGE startup’s AI-led intrusion
Cathedral, a cybersecurity startup founded by former DOGE staffers, says its AI agents found ways into Anthropic’s internal systems in less than 12 hours by abusing weaknesses in third-party business software, reaching meeting transcripts, salary and job offer data, and Slack discussions about the company’s most advanced models, according to a draft report reviewed by the Washington Post (paywalled). Anthropic, which received the findings through its bug bounty program, says only a limited set of sales, HR and IT systems was reached, that it cut off the startup’s access the same day, and that Claude conversations, source code and model weights were not affected. It also accuses Cathedral of breaking the bounty program’s rules by not disclosing all of its activity.
Midterm spending disclosures stall after FEC intrusion attempt
The US Federal Election Commission took its electronic filing system offline after someone attempted to breach its systems, according to an internal FEC email seen by Reuters, leaving political groups unable to submit campaign finance reports a week before the November 3 midterm elections. The outage blocked the 24-hour and 48-hour reports that track last-minute spending on specific races, although past filings remained searchable. The system was restored on October 28.
Hackers obtain national ID numbers, card PINs in University of Copenhagen breach
The University of Copenhagen says hackers who gained access to one of its test environments obtained personal data on about 525,000 current and former members of the university community, including names, phone numbers and personal email addresses, along with national identity (CPR) numbers for many of them. The attack was detected on October 20 and contained, but the PIN codes for the ID cards of roughly 20,000 staff and students were also compromised and have since been reissued. The university says it does not yet know who is behind the attack or what the motive was.
ENGIE data breach traced to tool used by sales partners
French energy giant ENGIE says attackers stole customer data after gaining access to a tool its partners use for customer prospecting and marketing energy offers, but that its own internal IT systems were not compromised. The incident is limited to customers in France, and notification letters say the stolen data includes names, contact details, customer IDs, contract numbers and energy supply point identifiers, but no banking information. ENGIE has not said how many customers are affected, and no group has claimed the attack.
Fake ChatGPT and Claude bots probe sites for leaked secrets
Cequence has tracked a single threat actor that rotates through about 32 AI bot identities (including crawlers and assistants from OpenAI, Anthropic, Meta and Perplexity) while scanning websites for files that commonly hold secrets, such as .env files, cloud credentials and private SSH keys. Running from a major cloud provider since at least late September, the scanner used more than 300 IP addresses in a single day and, on one day in October, reached 48% of the Cequence-protected organizations that saw any AI bot traffic. Cequence warns that a bot’s self-declared identity can’t be trusted, noting that the impersonator accounted for more than 80% of traffic labeled as AI bot traffic at some organizations.
Cloudflare and SGLang fix cross-tenant data leak in AI inference
Cloudflare says that between August 27 and September 9, 155 responses from the GLM 5.3 model on its Workers AI platform contained data from other customers’ recent requests to the same model, all involving prompts longer than 681,280 tokens. The leak resulted from a configuration mismatch that left the model’s KV cache smaller than its advertised context window, combined with a bug in the SGLang inference engine that allowed over-length requests to read cache data left behind by earlier sessions. Cloudflare fixed the issue after Knostic researcher Gadi Evron reported it via HackerOne, found no evidence of deliberate exploitation, and recommends that SGLang operators update to version 0.5.22 or later.
Defense worker, Russian money movers targeted over North Korean IT scheme
John Allen Kurrle Jr., 62, a defense contractor employee with a Top Secret clearance, has pleaded guilty to wire fraud conspiracy for aiding a North Korean remote IT worker scheme by letting foreign individuals use his identity (and that of a third person) to apply for remote jobs and hosting victim companies’ laptops at his home. The applications led to nine job offers from US companies, and Kurrle faces up to 20 years in prison. Separately, OFAC sanctioned 10 individuals and one entity supporting North Korea’s IT worker schemes, including two Russians who allegedly converted the workers’ crypto earnings into cash, two North Koreans who managed Russia-based IT worker groups, and the Russia-based First Financial and Technical Development Center (FFTDC), which deploys North Korean IT workers abroad.
Rockwell aims SecureOT at OEMs facing CRA and machinery rules
Rockwell Automation has added OEM-focused capabilities to its SecureOT offering to help machine builders prepare for cybersecurity requirements under the EU’s Cyber Resilience Act (CRA) and Machinery Regulation. Built on the SecureOT Platform, the additions include machine-level inventories of hardware, software and firmware, vulnerability identification, continuous monitoring for newly disclosed flaws after deployment, and exportable compliance reports for customers and auditors. Rockwell is also offering assessments and advisory services to help OEMs find gaps, set up governance processes, and align with standards such as IEC 62443.
AI moves deeper into ransomware, C2 malware and grey-market tooling
Sophos says the Gold Ogre group behind Aur0ra ransomware has used AI extensively, with agents orchestrated through Cursor handling network discovery, credential attacks and lateral movement with limited human involvement, after the operators bypassed guardrails by framing the intrusions as authorized red team work. Cisco Talos has detailed NIK C2, a new AI-integrated command-and-control framework whose Windows implant can steal credentials, stream keystrokes and deploy screen-locking ransomware. Symantec and Carbon Black, meanwhile, warn of a grey-market relay service, seen at a dozen organizations mostly in mainland China, that installs the genuine Codex and Claude Code CLIs but reconfigures them to route all requests through third-party servers with sandboxing disabled.
Related: In Other News: ICS-Sniper Attack, Hacking XWorm Malware Operators, $250k Chrome Exploit
