Daily Cyber Threat Brief — October 31, 2026: Chainalysis breach blamed on 'offensive agentic techniques'

What "offensive agentic techniques" means here is not yet established. Chainalysis has shared no technical details: not the vulnerabilities exploited, not which internal tools were accessed, and not the dwell time. BleepingComputer notes it is unclear whether the attackers used an existing agentic offensive framework such as ARTEX — recently used by threat actors in the ConnectWise supply chain compromise, according to Microsoft — or custom-built AI agents able to identify vulnerabilities, exploit them, and chain subsequent actions with limited human intervention. The disclosure is explicit about what was not touched: no evidence of access to investigation content, authentication credentials, or session tokens, and no connection to the Binance theft. What remains open: the number of affected customers (Chainalysis says it is still determining which records were accessed for each organization), and whether the exposed contact and role data enables targeted phishing against its law-enforcement and exchange customers — the exact risk its own warning about impersonation and fund-recovery scams points to.

The figures count different things and cannot be added into a single passenger total: Cathay reports bookings, JAL and ANA report customers, and no carrier has published a complete industry-wide count. The access window — August 10, 2023 to September 23, 2026 — means the exposure ran for more than three years before detection. JAL says credit card numbers and passwords were not compromised, that Amadeus has completed the necessary security measures, that there is currently no evidence of misuse, and that flight operations are unaffected. The pattern is a third-party breach at scale: a single supplier compromise reaching tens of thousands of people at each carrier without touching the airlines' own networks. Amadeus has not said how many of its airline customers were hit, and investigations continue.

The GitLab case carries a patching trap worth flagging: Alpine Linux's October 28 post-mortem showed an attacker tried the flaw against Alpine's own GitLab server, and upgrading GitLab alone did not close the hole on a standard-Ruby install — the Ruby update (versions 3.3.13, 3.4.12, and 4.0.9, released October 23) was also required for installs running GitLab on their own Ruby. The Cisco flaw is the second Cisco VPN zero-day since August, and its effect — forcing a firewall reload on request — is a denial-of-service primitive against the box that terminates remote access. Drupal separately warned that CVE-2026-118539, a missing-authorization flaw in Commerce Core, is under active attack. The official anchor for all four KEV entries is CISA's catalog listing and the November 1 federal deadline; treat any of these as a patch-now queue item, not a routine update.

Scope this precisely. Remote code execution has been demonstrated only on servers with an MPTCP listener; the flaw also affects ordinary TCP, but remote code execution has not been publicly demonstrated on servers without an MPTCP listener, and there are no confirmed in-the-wild attacks. That distinction matters because the surface may be wider than it looks: BleepingComputer reports that Go 1.24 and later can enable MPTCP by default for Linux listeners when the kernel supports it, so a service owner may need to check application behavior rather than assume the feature was intentionally configured. The fix is in the upstream kernel and maintained branches, with distribution availability varying — Debian has shipped it and AlmaLinux has patched kernels in testing. Disabling MPTCP alone is not sufficient, because ordinary TCP can still trigger the flaw; patching the kernel is the remediation.

The two intrusions diverged at the certificate layer. In the first, attackers found a certificate template vulnerable to ESC1 — broad enrollment permissions, requester-supplied identity information, and weak issuance controls — letting a low-privileged account request authentication certificates for privileged identities. The first phishing wave targeted three users and compromised one workstation, yielding certificates for three highly privileged accounts; about 90 minutes after disruption, a second wave targeted 16 users, compromised another workstation, and secured certificates for eight more privileged accounts. One privileged identity excluded from automated containment later supported DCSync and data theft, while Microsoft Defender blocked more than 1,000 subsequent SMB file-access attempts and additional logons. In the second intrusion, attackers used PetitPotam over EFSRPC to coerce domain-controller authentication and relayed the NTLM authentication to inadequately protected AD CS Web Enrollment, exploiting the ESC8 weakness to obtain a domain-controller certificate and a ticket-granting ticket via Kerberos PKINIT; the customer isolated the workstation and no further progression was observed. The durable lesson: a password reset does not revoke an issued certificate, so AD CS template permissions, web-enrollment protections (HTTPS and Extended Protection for Authentication), and containment exclusions all need review — not just the help-desk workflow that let the attacker in.