Today's PaperCut and JFrog Artifactory activity extends several threads already seen this week, but both incidents show attackers making greater use of complete exploit chains rather than relying on a single flaw. In JFrog Artifactory, attackers combined CVE-2026-42018, CVE-2026-42016 and CVE-2026-82329 to move from anonymous access to administrative control before installing malicious plugins and a Rust-based backdoor. PaperCut followed a similar progression: an exploit was developed and tested in a private environment before AI agents deployed it against 395 organisations, with one intrusion reaching domain administrator in under six hours. Neither campaign relied on a novel technique. Once a working chain had been established, it could be reused across many targets with relatively little additional effort.

A separate set of breaches involved access originating outside the organisation ultimately affected. Brevo's SAML SSO flaw gave an attacker access to 138 customer accounts, six of which were then used to phish 347,000 Trezor, BitBox and CoinTracking users. Florida's DMV breach instead involved police credentials stored on an officer's personal device. In both cases, the initial weakness sat outside the organisation whose users or systems were ultimately exposed. This is similar to the intermediary-data breaches seen with IDScan and Thomson Reuters earlier this week, but the exposure here came through inherited access and trust rather than data held by the intermediary itself.

Oleksii Lytvynenko, 44, was sentenced to four years in prison for wire fraud conspiracy tied to the Conti ransomware operation. Between 2020 and 2022, Conti infected over 1,000 victims across 47 US states and 31 countries, with payouts exceeding $150 million by January 2022. Lytvynenko admitted possessing stolen data from twelve victims and coding a loader used to deploy further malicious payloads.

GitLab patched CVE-2026-85706, a maximum-severity path traversal vulnerability in the repository commits API allowing unauthenticated attackers to read arbitrary data. Security firm watchTowr reported attackers already scanning for unpatched, internet-exposed GitLab servers a day after disclosure. GitLab also fixed a second critical flaw, CVE-2026-87719, an insecure deserialisation weakness in the GraphQL subscription serializer affecting GitLab EE.

Anthropic disrupted a cyberespionage operation attributed with high confidence to Russian state-nexus group Midnight Blizzard, tracked internally as GTG-20006. The actor used AI agents to monitor whether security products flagged its malware, then automatically modified and redeployed the tools until they evaded detection again. Targets included over 20 government, defense, diplomatic and think-tank organisations across Ukraine, Europe, the Middle East and Asia, alongside theft of a drone vision system's proprietary software development kit and takeover of victims' WhatsApp accounts via linked companion devices.

JFrog disclosed CVE-2026-82329, a critical authentication weakness in Artifactory that could let an unauthenticated attacker with network access gain administrative privileges. The flaw affects multiple release branches under default configuration and has been fixed in versions 7.111.21 through 7.161.20. Cloud environments have already been remediated, while self-hosted deployments require upgrading or applying a join-key workaround.

Sophos researchers identified a new variant of the Cyclops Blink malware on compromised Cisco Firewall Management Center devices, linked with moderate confidence to Russia's Sandworm group. Unlike the 2022 version targeting WatchGuard firmware, the 2026 variant runs on x86-64 Linux and adds active network scanning, selective packet capture, and generic persistence via a fake SysV init service. The implant communicates over TLS with a hard-coded command-and-control server and can download, execute, or register new payloads as additional modules.