Daily Cyber Threat Brief — October 28, 2026: GMO customers told to power down as attackers publish stolen data

Telling customers to power down their own machines is a last-resort containment step, and it signals that GMO believed the attacker still held a live foothold with a ransomware payload ready to fire. The sequence matters: the intrusion was detected October 20, the service was shut down October 21, and the extortion escalated a week later when the attacker set a noon October 27 deadline to publish customer data and trigger ransomware. GMO confirmed the publication in a report on October 27 and said it was checking whether the leaked material is genuine.

Two details separate confirmed damage from claimed damage. First, GMO says payment information was not exposed because it sits on separate infrastructure — but MetaTrader 4 and 5 credentials and trading-account details may have been exposed, and individual customers have reported unauthorised trades and losses. Second, the ransomware threat itself appears not to have materialised: no ransomware was activated on customer machines, according to BleepingComputer's account. That does not clear the machines — GMO is still scanning customer devices for malware, and customers whose information was exposed face follow-on fraud risk from the published data. GMO has said it will not comply with the demands, offered ¥5,000 (about $30) per customer in compensation, and warned against redistributing the published information.

Warp, the agentic development environment used by more than 800,000 developers, disclosed that attackers exploited a third-party vendor in its infrastructure and exposed agent decision logs and user telemetry. Founder and CEO Zach Lloyd said the company became aware on Friday of what it believes was "an active and narrow-banded exploitation of a vendor in Warp's infrastructure," detected the operation within hours, and estimated that "roughly 1% of Warp users may have been impacted." CrowdStrike is assisting the investigation and law enforcement has been informed; affected users have been notified.

The unresolved question is the sensitivity of what was taken. Agent traces and telemetry were identified as potentially affected, but Warp has not clarified whether source code, prompts, credentials or access keys were included. That distinction is the whole story for developers: traces can reveal how agents were instructed and what they touched, which is a different risk profile from raw telemetry. Risky Business's account, cited in the reporting, describes the intrusion as powered by "an agent swarm" using "a very sophisticated penetration and escalation technique" — if that characterisation holds, it is a notable escalation in its own right: AI agents used as the intrusion tooling against an AI developer platform. The incident follows a breach disclosed last week at AI coding platform OpenCode, in which stored API keys for external providers were exposed. The pattern to watch is the supply chain around AI tooling, not just the tools themselves.

The University of New Mexico Health Sciences Center detected a cyberattack on October 26 that forced it to divert ambulances for about two hours, limit new patient admissions, suspend outpatient surgeries and cancel appointments, and cancel classes at its four health sciences schools. UNM Hospital in Albuquerque is New Mexico's only Level I trauma centre, which is why a two-hour diversion matters beyond the hospital itself. By October 27, UNM said critical systems were back online, hospitals were fully open, ambulance traffic was normal, and clinics had resumed appointments.

The new ransomware group Byod claimed the attack, saying it compromised 36,000 systems across UNM's main campus and hospital environments, including Active Directory forests, the entire Microsoft Entra tenant and two Oracle databases, and accessed Social Security numbers, payroll records, research data and patient information. That claim is unverified, and UNM's own assessment points the other way: it found some end-user devices affected but no penetration of core systems or servers. Treat the two accounts as a live discrepancy, not a settled fact — the confirmed impact is the care disruption on October 26, while the scale of any data theft rests on an unproven extortion claim. UNM has advised patients, staff and students to watch for unsolicited communications and phishing.

Attackers have begun scanning the internet for systems unpatched against CVE-2026-101767, a critical unauthenticated SQL injection in Oracle E-Business Suite's Thin Client Framework servlet that chains to remote code execution. Oracle addressed the flaw in its October 20 Critical Patch Update and urged customers to patch immediately; Oracle has not tagged the flaw as exploited in the wild, but threat intelligence firm Defused said its Oracle E-Business Suite decoys recorded exploratory exploits against the CVE on October 27. Shadowserver tracks more than 670 internet-exposed Oracle EBS instances, most with US IP addresses, though that count does not distinguish honeypots from genuinely vulnerable systems.

The window between patch and probe here was seven days, and Oracle's own warning — that attackers have succeeded in some instances because targeted customers had failed to apply available patches — is the operational point. Oracle EBS sits inside the enterprise systems of large corporations, government organisations and healthcare networks, which is why its vulnerabilities draw immediate scanning. Any internet-facing EBS instance still unpatched should be treated as a priority-one patch, not a routine maintenance item.

Denver Health Medical Plan is notifying 238,083 members that their personal and health information may have been compromised after a third-party vendor, SquareML, inadvertently published an access key on a publicly accessible website on June 27, 2025. SquareML notified the health plan on August 14, 2026, and the key was revoked on August 18, 2026 — leaving a window of more than a year in which the key could have been used to reach a SquareML database holding member data.

The exposure is bounded in two ways that matter: the account was read-only, so the key could not download, export, delete or change information, and the database held no Social Security numbers or financial information. What it did hold is still sensitive: names, dates of birth, ZIP codes, demographic information, health metrics, plan eligibility data, hospital admission, provider name, and discharge and transfer data. Denver Health Medical Plan says it has no evidence the information was viewed or misused, only that it may have been. The lesson is a familiar one with an unusually long fuse: a leaked credential is not a single incident but an exposure window, and the window here ran for over thirteen months before revocation.