Daily Cyber Threat Brief — November 3, 2026: Harvard breach exposed 851,000 accounts

Incident timeline

Date	Development
August 2025	Clop ransomware gang steals data from nearly 3.5 million University of Phoenix students, staff and suppliers via an Oracle E-Business Suite zero-day (revealed two months after Harvard's October disclosure).
October 2025	Harvard tells BleepingComputer it is investigating a second breach after Clop claims its systems using the same Oracle EBS flaw.
November 18, 2025	Harvard discovers the Alumni Affairs and Development breach; Princeton and the University of Pennsylvania disclose donor-data breaches the same month.
Early February 2026	ShinyHunters leaks a 1.1GB archive of Harvard documents on its dark web leak site after the university refuses to pay.
October 29, 2026	CISA adds GitLab CVE-2026-97229 and Cisco CVE-2026-76418 to the Known Exploited Vulnerabilities catalog, with a November 1 remediation deadline for federal agencies.
October 30, 2026	Chainalysis discloses an AI-assisted intrusion into its systems and internal tools.
November 2, 2026	Google announces the XCPE botnet disruption; Shadowserver reports more than 1,000 GitLab servers still unpatched against the exploited RCE flaw.
November 3, 2026	Have I Been Pwned confirms the Harvard breach exposed 850,500 accounts.

1. Harvard breach exposed 851,000 accounts

Have I Been Pwned's analysis of the data ShinyHunters leaked puts a firm number on a breach Harvard disclosed nearly a year ago: 851,000 unique email addresses, with extensive profile information for donors, alumni, parents and some faculty, staff and students. The data spans names, dates of birth, phone numbers, home and business addresses, social media profiles, donation details and event attendance records. Harvard's chief information officer, Klara Jelinkova, said the breached systems stored no Social Security numbers, passwords, payment card information or financial information — a meaningful limit on the damage, but the donor and alumni profile data is exactly the material used for convincing impersonation and fundraising-themed phishing.

The context matters: Harvard is the third Ivy League school hit in the same wave — Princeton and the University of Pennsylvania both disclosed donor-data breaches in November 2025 — and the university is separately dealing with a Clop claim tied to an Oracle E-Business Suite zero-day, the same flaw Clop used against the University of Phoenix, where nearly 3.5 million records were taken. Universities' alumni and advancement systems remain a soft, high-value target: rich contact and wealth-indicator data, often on older platforms.

Investigation notes

Attribution: Harvard has not attributed the attack; ShinyHunters' claim and the February leak are the public evidence. The 850,500-account figure comes from Have I Been Pwned's analysis of the leaked dataset, not from Harvard — the university declined to share an affected-individual count when BleepingComputer asked at the time. The ShinyHunters claim of "over 1 million records" is a claim, not a confirmed count; the confirmed figure is the HIBP number. No evidence has been published that Social Security numbers or credentials were involved.

2. Google dismantles 1.3-million-device XCPE botnet, bans the TV box maker

Google and a coalition of security firms and law enforcement disrupted XCPE, a botnet built from more than 1.3 million Android TV streaming boxes. Google's Threat Intelligence Group says, with high confidence, that the backdoor was deliberately built, deployed and operated for financial gain by Shenzhen-based Videostrong Technology Co., which sold access to criminal affiliates who ran their own schemes through victims' living rooms. A separate telemetry count from partner TrendAI put the figure at 1,472,554 infected devices across 206 models — a different measurement from Google's 1.3-million headline count, not a contradiction. The backdoor also turned up in other Android-based IoT devices, including telehealth devices, karaoke machines and digital signage.

Google has banned the maker, and the practical advice for owners is blunt: a standard antivirus scan or deleting a suspicious app will not remove a firmware-level backdoor. The realistic options are installing clean firmware from a trustworthy source or disconnecting the device from the internet.

Investigation notes

The deliberate-backdoor claim is Google's assessment, corroborated by an independent firm's investigation — not a court ruling — and Videostrong did not respond to press inquiries. The distinction matters: this was not a missed vulnerability but, per Google, an intentionally shipped backdoor. The takedown shut down the botnet's servers in late October, with the announcement following on November 2.

3. AI agents run a 200-target espionage campaign: Vapor Panda and ARTEX

CrowdStrike's reporting on the China-nexus group it tracks as Vapor Panda shows how far agentic attacks have scaled: starting October 17, AI agents built on the ARTEX multi-agent penetration-testing framework targeted more than 200 organizations across 14 countries in 11 days, directed by a human operator issuing instructions in Chinese and intervening at decision points. Confirmed victims include telecommunications providers in the United States and Belgium, government organizations in Greece, the Middle East and Latin America, and pharmaceutical firms in the U.S. CrowdStrike confirmed post-exploitation activity at nine organizations — where agents forged an administrator token and read mailboxes in the Greek government intrusion, and deployed the BLOODALCHEMY backdoor against government entities in Latin America and the Middle East — plus password spraying and credential scanning against at least 12 more high-value targets. In one intrusion, the planning agents consumed more than 1.4 billion input tokens.

ARTEX began as an open-source project by Chinese developers and was later made proprietary after its use in real-world attacks, including a campaign against South Korean financial institutions. Microsoft has separately attributed ARTEX activity to Storm-0940. The campaign is a marker for defenders: one operator can now run simultaneous engagements that once needed a full team.

Investigation notes

Do not conflate "200 targeted" with "200 breached": CrowdStrike confirmed post-exploitation at nine organizations, with about 70% of targets assessed as confirmed compromised or showing moderate-confidence evidence of intrusion, per CrowdStrike's comment to CyberScoop. ARTEX ran mainly on DeepSeek plus a second, unidentified commercial model. The framework's developer withdrew the open-source code in October after its misuse. This story pairs with the Chainalysis disclosure below: agentic techniques are now showing up on both the espionage and the criminal sides of the ledger.

4. Chainalysis breach: attacker used "offensive agentic techniques"

Blockchain analytics firm Chainalysis disclosed on October 30 that an external actor using "offensive agentic techniques" gained unauthorized access to its systems and internal tools, reaching user IDs, full names, email addresses and organization role labels for users provisioned on its platform. The company says it has found no indication that investigation content — graphs, notes or cases — transaction or query results, clear-text passwords, MFA secrets or session tokens were accessed, and it stresses the incident is unrelated to the Binance theft it is helping trace. The number of affected customers, the entry point and how the AI agents were used have not been disclosed.

The risk here is not the sensitivity of any single record but the directory it creates: names, work emails and role labels identifying who handles investigations and compliance, and at which organization — precise raw material for targeted phishing against the firms that trace crypto crime.

Investigation notes

Chainalysis has not named the attacker, the exploited vulnerability or the AI model involved, and declined follow-up questions from The Block. The disclosure was published as a customer notice on October 30; the company says it is still determining which records belong to each customer and will send per-organization breakdowns. Treat the "offensive agentic techniques" label as the company's characterization, not an independently verified technical description.

5. More than 1,000 GitLab servers still open to the exploited RCE flaw

Three weeks after GitLab patched CVE-2026-97229 — a maximum-severity integer overflow in the CI/CD regular expression compiler that lets an unauthenticated attacker run code on the server via a crafted regular expression in a pipeline configuration — Shadowserver reports more than 1,000 internet-exposed GitLab instances still vulnerable. CISA added the flaw to its Known Exploited Vulnerabilities catalog on October 29, with a November 1 remediation deadline for federal civilian agencies. Shadowserver's same-day telemetry also counted more than 11,300 instances vulnerable to a second GitLab flaw, CVE-2026-110264, among roughly 39,000 GitLab instances it sees daily. Fixes are in GitLab 19.4.2, 19.3.5 and 19.2.8; GitLab.com was already patched.

Investigation notes

The KEV listing is the official anchor for active exploitation here — CISA's catalog lists only flaws with evidence of real-world exploitation. The exposure counts are Shadowserver telemetry, not confirmed compromises: "vulnerable" does not mean "breached." Alpine Linux's self-hosted GitLab saw attempted exploitation in which crafted requests crashed worker processes, with no evidence of successful code execution. Self-managed GitLab patching lags are the recurring lesson: the patch has existed since October 12.

Sources

• BleepingComputer — Harvard University data breach exposed info of 851,000 accounts
• IT Security News — Google dismantles 1.3 million device XCPE botnet, bans Android TV maker
• BleepingComputer — Chinese hackers use ARTEX AI in automated attacks against 200 orgs
• The Block — Chainalysis says hacker using 'offensive agentic techniques' accessed its customers' names and emails
• Security Affairs — Chainalysis investigates security incident involving AI-driven attack techniques
• BleepingComputer — Over 1,000 GitLab servers vulnerable to ongoing RCE attacks