CISA added CVE-2026-63077 to its Known Exploited Vulnerabilities catalog on August 5 and set a federal remediation deadline of August 8, confirming active exploitation nine days after JetBrains disclosed the flaw on July 28. CVE-2026-63077 is a deserialization of untrusted data vulnerability in TeamCity's agent polling protocol, the channel build agents use to request job assignments from the central server. An unauthenticated attacker with HTTP or HTTPS access to a TeamCity server can send a crafted payload to that endpoint, bypass authentication entirely, and execute arbitrary commands as the TeamCity server process. Every historical TeamCity On-Premises release is affected. Fixed versions are 2025.11.7 and 2026.1.3. JetBrains also provides a security patch plugin compatible with 2017.1 and later that addresses this CVE without a full version upgrade. Censys identified approximately 4,500 internet-accessible TeamCity web properties shortly after disclosure, a meaningful number given the prior exploitation history. Neither CISA nor JetBrains has published specific attribution for the current exploitation activity. The historical comparison is not reassuring: CVE-2023-42793, a critical authentication bypass in TeamCity, was exploited by North Korea's Diamond Sleet and Onyx Sleet within weeks of public disclosure and separately by Russia's SVR, with post-exploitation activity including malware deployment, credential dumping, and scheduled-task persistence. CVE-2024-27198 was added to CISA KEV only days after public disclosure. CVE-2026-63077 now follows that same trajectory, one week after this brief covered it as urgent in Issue 95 with no exploitation confirmed at that time.

Security researcher Asim Manizada published the full write-up and proof of concept for CVE-2026-64531 on the oss-security mailing list on July 28, 2026, following a coordinated disclosure process that began on June 19. The upstream kernel fix had already landed in stable branches on July 24, four days before the public disclosure. The flaw is an integer wraparound in the Open vSwitch kernel module's handling of internally generated Netlink action streams. Open vSwitch accepts a list of network actions from userspace and rewrites them into an internal format stored as Netlink attributes. The length field for each nested attribute is 16 bits wide, capping it at 65,535 bytes. When the total internal action stream grows past 64 KB across multiple nested levels, the length counter wraps around. The resulting write lands in memory past the allocated buffer, overwriting adjacent kernel structures with attacker-controlled content. The corruption is deterministic rather than probabilistic, giving the exploit a reliable primitive rather than a race. On systems where unprivileged user namespaces are enabled, an ordinary local account can create a private user and network namespace, obtain administrative capabilities inside that namespace, and trigger the overflow without any pre-existing Open vSwitch configuration. The published exploit includes pre-built heap spray records for roughly 800 x86-64 kernel builds and attempts dynamic derivation for additional configurations. The flaw is present in kernel code dating back 13 years. No exploitation in the wild has been confirmed. Distribution patches are shipping at different speeds: RHEL, AlmaLinux, and Debian are affected; patch availability varies by vendor and release track.

Researchers documented a campaign by a Chinese-speaking threat actor operating under the aliases knaithe and KnYuan in which DeepSeek, running through the Hermes Agent framework, served as an autonomous offensive operator targeting internet-exposed web infrastructure. The agent was tasked with exploiting vulnerable targets and directed the full attack cycle: initial reconnaissance against exposed services, selection of relevant CVEs, exploitation attempts, result evaluation, and adaptation when a chosen technique failed. When initial attempts to exploit a Langflow vulnerability, CVE-2026-33017, were blocked by the target environment's restrictive configuration, the agent autonomously switched to CVE-2026-34486, an Apache Tomcat vulnerability that bypasses the EncryptInterceptor cluster component. The campaign ultimately breached 107 endpoints and included 16 root-level compromises of cPanel and WHM servers via CVE-2026-41940. CISA added CVE-2026-9198, a code injection vulnerability in Langflow allowing unauthenticated remote code execution on default deployments, and CVE-2026-34486 to its KEV catalog on August 5 in connection with confirmed active exploitation. The attack chain used publicly available GitHub proof-of-concept exploit code, a single pirated commercial command-and-control product, and one tracked named malware family. No novel tooling was required; the AI system handled target selection, technique selection, and adaptation.