This Week in Cybersecurity
Issue 18 is dominated by a single recurring adversary and a sobering reminder that the attack surface for AI infrastructure has quietly become as dangerous as anything in the enterprise stack.
ShinyHunters had a prolific week. The group confirmed responsibility for two major breaches in the span of 48 hours — Medtronic (9 million claimed records, now confirmed by the company) and ADT (5.5 million customers, verified by Have I Been Pwned). Taken together, these two incidents add tens of millions of individuals to ShinyHunters' already enormous 2026 victim roster. The healthcare angle at Medtronic is particularly serious: depending on what the investigation turns up, HIPAA breach obligations at scale, potential exposure of device-adjacent data, and regulatory obligations across 150+ countries could all come into play simultaneously.
On the vulnerability front, the week delivered two critical flaws that deserve immediate attention. CVE-2026-3854 is a CVSS 8.7 remote code execution flaw in GitHub and GitHub Enterprise Server, exploitable with a single authenticated git push — no special privileges, no target interaction, no complex prerequisites. GitHub.com is patched; Enterprise Server customers must act manually. Meanwhile, CVE-2026-42208, a pre-authentication SQL injection in LiteLLM, is already being actively exploited in the wild. LiteLLM sits at the center of many enterprise AI pipelines, storing API keys for OpenAI, Anthropic, Azure, and others — making a successful injection attack a skeleton key for an organization's entire AI infrastructure.
And in what may be the most unusual story of the week: rival ransomware groups 0APT and KryBit turned their operations on each other in a dispute over revenue, attribution, and stolen tooling. The mutual leak of C2 infrastructure, victim lists, and internal tooling gives defenders a rare and direct window into active gang operations — treat the exposed indicators as priority intelligence.
Top Stories
Medtronic Confirms Breach After ShinyHunters Threatens 9 Million Record Leak
Medical device giant Medtronic has officially confirmed it suffered a cybersecurity breach after ShinyHunters threatened to publicly release data claimed to represent 9 million individuals. The company had previously disclosed unauthorized access to "certain corporate IT systems" but had not confirmed the scope — ShinyHunters' extortion deadline forced a more definitive position.
The severity here goes beyond a standard enterprise PII breach. Medtronic manufactures implantable cardiac defibrillators, insulin delivery systems, surgical robotics, and spinal stimulation devices. Depending on the systems touched, the breach could involve not only employee and customer PII but also device-adjacent clinical and regulatory data. If Protected Health Information is confirmed, HIPAA obligations kick in at scale — mandatory notification for potentially millions of individuals, HHS reporting within 60 days, and regulatory exposure across the dozens of jurisdictions where Medtronic operates globally under GDPR, PIPEDA, and national health data laws.
ShinyHunters has a documented history of following through on leak threats when demands go unmet. Affected individuals should watch for official breach notifications, freeze credit at all three major bureaus, monitor healthcare EOB statements for unfamiliar claims, and be especially alert to impersonation attempts from callers or emails claiming to be Medtronic.
Critical GitHub CVE-2026-3854 RCE Flaw — One Git Push Is All It Takes
Researchers have publicly disclosed CVE-2026-3854, a critical remote code execution vulnerability affecting both GitHub.com and GitHub Enterprise Server, carrying a CVSS score of 8.7. The flaw lives in GitHub's server-side git push processing pipeline: a specially crafted git object, pushed via a standard git push command by any authenticated user, is sufficient to trigger code execution on the server — no elevated privileges, no prior reconnaissance, no complex payload delivery chain.
GitHub.com has been patched. The risk is immediate and specific to GitHub Enterprise Server deployments, where administrators must manually apply the available hotpatch. The supply chain implications are substantial: an exploited GHES instance exposes private repository contents, CI/CD pipeline secrets, and environment variables — and the attack requires nothing more than a free GitHub account and a crafted commit. This follows a recognized pattern of git protocol server-side processing vulnerabilities (CVE-2022-24765, CVE-2023-22490, CVE-2024-32002) — each demonstrating that the git object model's complexity makes it an enduring attack surface.
GHES administrators should apply the patch immediately, review audit logs for suspicious push activity from the pre-patch window, rotate any secrets stored in repository environment variables, and enable audit log streaming for real-time visibility.
LiteLLM Pre-Auth SQL Injection CVE-2026-42208 Under Active Exploitation
Threat actors are actively exploiting CVE-2026-42208, a critical pre-authentication SQL injection flaw in LiteLLM, the popular open-source LLM gateway used by enterprises to route API calls across OpenAI, Anthropic, Azure OpenAI, Google Gemini, and AWS Bedrock. Because LiteLLM stores API keys, routing configurations, budget settings, and usage logs for every connected provider, a successful injection against an exposed instance can hand an attacker a complete skeleton key to an organization's AI infrastructure.
The pre-authentication classification is the critical differentiator: exploiters do not need a valid account. Any internet-exposed LiteLLM instance is a target. Active campaigns are using the extracted credentials to drain API quotas, exfiltrate historical prompt data, and pivot directly into upstream provider accounts. Researchers note this fits a broader trend — AI infrastructure is now a primary attack surface, following the Vercel/Context AI breach (Issue 17) and similar incidents targeting centralized AI tooling. Organizations running LiteLLM should update immediately, rotate all stored API keys, audit OpenAI/Anthropic/Azure usage dashboards for anomalous spikes, and restrict LiteLLM access to internal networks or VPN.
ADT Data Breach: ShinyHunters Exposes 5.5 Million Customers
Home security giant ADT has confirmed a data breach affecting 5.5 million individuals — independently verified by Have I Been Pwned, which has added the dataset to its database. The breach, attributed to ShinyHunters, exposed full names, email addresses, phone numbers, and physical home addresses of ADT residential customers across the United States.
The combination of exposed data carries layered risk for ADT customers. Names, emails, and phone numbers enable convincing phishing and vishing campaigns; physical addresses combined with ADT enrollment status could theoretically inform physical targeting. The breach is ShinyHunters' second confirmed high-profile victim this week alone, cementing the group's status as 2026's most prolific data extortion threat actor. ADT customers should check Have I Been Pwned, enable two-factor authentication on their ADT account, freeze credit at all three bureaus, and contact their mobile carrier to add a SIM-swap protection PIN.
Feuding Ransomware Groups 0APT and KryBit Leak Each Other's Infrastructure
In a rare development in the cybercriminal ecosystem, rival ransomware groups 0APT and KryBit have turned their operations against each other — the result of a dispute over revenue sharing, victim attribution conflicts, and stolen tooling accusations. The mutual intrusion and public data dump has produced an unprecedented intelligence windfall for defenders: active C2 server addresses, operational domain lists, victim targeting criteria, affiliate communications, and the specific tooling each group uses for lateral movement and credential theft.
The leaked targeting data reveals how both groups prioritize victims — revenue thresholds, sector weighting (healthcare, finance, and manufacturing rated as premium), and pre-intrusion Shodan/Censys scanning for exposed services. Security teams should immediately ingest the exposed indicators into SIEM and threat intelligence platforms, scan historical firewall logs for C2 connections matching the leaked infrastructure, and update EDR detection rules for the exposed TTP signatures. This follows a documented historical pattern: Conti's 2022 internal chat leak, REvil's 2021 source code exposure, and LockBit's 2023 builder leak each produced similar defender windfalls from criminal infighting.
Security Corner
10 CVEs are newly published to the Security Advisories section this week. Key advisories to prioritize:
CVE-2026-7224 — SQL Injection in SourceCodester Pizzafy Ecommerce System 1.0 (CVSS 7.3)
A high-severity SQL injection vulnerability in the /admin/ajax.php delete_cart handler allows remote attackers to enumerate the database, extract customer PII, and harvest admin password hashes. No official patch is available — restrict admin panel access to trusted IPs immediately and replace direct SQL concatenation with prepared statements.
Full advisory →
CVE-2026-7154 — High-Severity Vulnerability A high-severity flaw disclosed this week affecting enterprise deployments. Review the advisory for affected versions and apply available patches. Full advisory →
CVE-2026-7136 — High-Severity Vulnerability Additional high-severity disclosure this week. Organizations should assess applicability against their environment and patch on an accelerated schedule. Full advisory →
Also published this week:
- CVE-2024-46636 → Advisory →
- CVE-2026-41462 → Advisory →
- CVE-2026-30352 → Advisory →
- CVE-2026-41635 → Advisory →
- CVE-2026-40860 → Advisory →
- CVE-2026-40453 → Advisory →
- CVE-2026-7077 → Advisory →
Quick Takes
-
Broken VECT-2.0 Ransomware Acts as Data Wiper: A new ransomware variant called Broken VECT-2.0 has been found to function as a destructive wiper rather than an encryptor for files above a size threshold — meaning there is no recovery path even if a ransom is paid. Organizations should ensure offline, air-gapped backups are current and regularly tested, as wiper-hybrid ransomware removes the negotiation option entirely. Read more →
-
Vimeo Confirms Anodot Breach Exposed User Data: Vimeo has confirmed that user data was exposed following a breach at analytics partner Anodot. The incident highlights the continuing risk of third-party data processor relationships — Vimeo customers whose data flowed through Anodot should watch for direct notification and be alert to phishing targeting their Vimeo account credentials. Read more →
-
PhantomCore Exploits TrueConf Vulnerabilities to Breach Russian Networks: The PhantomCore threat actor group has been observed exploiting vulnerabilities in the TrueConf video conferencing platform to gain initial access to Russian enterprise networks. The campaign underscores that video conferencing software — expanded rapidly during the remote work era — remains an underpatched attack surface for both espionage and financially motivated actors. Read more →
-
Deepfake Voice Attacks Are Outpacing Defenses: A new analysis warns that AI-generated deepfake voice attacks are now sophisticated enough to defeat most current vishing defenses, including voice biometric verification systems. Organizations relying on voice authentication for financial approvals or executive wire transfer authorization should add out-of-band verification steps that cannot be replicated by audio synthesis. Read more →
-
Firefox Vulnerability Allows Tor User Fingerprinting: A vulnerability in Firefox enables fingerprinting of users routing traffic through Tor, potentially deanonymizing individuals who rely on the Tor Browser for privacy-sensitive activity. The Tor Project and Mozilla have been notified; users requiring strong anonymity guarantees should apply updates as they are released. Read more →
-
FTC: Americans Lost Over $2.1 Billion to Social Media Scams in 2025: The FTC has published its annual fraud report showing social media scam losses hit $2.1 billion in 2025, driven by investment fraud, fake marketplace listings, and romance scams. The figures reinforce that social engineering at scale remains one of the highest-impact and lowest-cost attack vectors available to threat actors. Read more →
-
GopherWhisper Go Backdoor Infects 12 Mongolian Government Systems: China-linked threat actor GopherWhisper has been confirmed to have compromised 12 Mongolian government systems using a Go-language backdoor. The campaign is consistent with ongoing Chinese APT interest in Central Asian government networks for intelligence collection. Read more →
Upcoming
-
Medtronic Breach Scope Determination: The critical unknown is whether Medtronic's breach involved Protected Health Information. If PHI is confirmed, mandatory HIPAA notifications must go out within 60 days of breach discovery — potentially to millions of individuals. Watch for formal breach notification filings with HHS and state attorney general offices in the coming weeks, which will give a clearer picture of actual scope.
-
GitHub Enterprise Server Patch Compliance Window: CVE-2026-3854 is patched on GitHub.com but requires manual action on self-hosted Enterprise Server deployments. The attack complexity is low and authentication requirements are minimal — organizations running unpatched GHES should treat this as a zero-hour remediation item and not defer to the next maintenance window. Check
ghe-versionoutput against the patched release to confirm status. -
LiteLLM Credential Rotation: If your organization runs LiteLLM, active exploitation of CVE-2026-42208 means any internet-exposed deployment should be assumed compromised until rotations are complete. Prioritize OpenAI, Anthropic, and Azure credentials first — these carry the highest immediate blast radius from quota abuse and data exposure. Revoke and reissue, then restrict network access before re-enabling.
-
0APT/KryBit IOC Operationalization: The leaked infrastructure data from the ransomware gang war is time-sensitive — C2 addresses and operational domains are most useful while the groups are still active on the same infrastructure. Threat intelligence teams should prioritize ingesting these indicators before the adversaries rotate their infrastructure in response to the exposure.
-
Vimeo/Anodot Third-Party Audit: The Vimeo breach via Anodot is a reminder to audit active data processor relationships. Organizations handling significant customer data should review their vendor inventory to identify which third-party analytics, monitoring, or BI tools have access to customer records — and confirm those vendors have appropriate security controls and breach notification obligations contractually defined.
By the Numbers
| Metric | Value |
|---|---|
| Medtronic breach — records claimed by ShinyHunters | 9 million |
| ADT breach — customers confirmed exposed | 5.5 million |
| CVE-2026-3854 (GitHub RCE) CVSS score | 8.7 High |
| CVE-2026-42208 (LiteLLM SQLi) authentication required | None — pre-auth |
| 0APT/KryBit leaked categories | C2 infra, victim lists, tooling, affiliate comms |
| FTC-reported social media fraud losses (2025) | $2.1 billion |
| Mongolian government systems breached by GopherWhisper | 12 |
| New CVEs published this week | 10 |
CosmicBytez Labs — IT & Cybersecurity Intelligence Hub