Cybersecurity briefing — July 9, 2026.

ESET’s H1 2026 threat landscape analysis highlights a continuing shift toward AI-enabled abuse paths, including malicious “AI skills” designed to interact with tools and data. The report notes attackers are leveraging agent capabilities to steal information, execute malware, and manipulate agent behavior—often by slipping harmful logic into seemingly legitimate workflow components. The practical takeaway for defenders: agentic systems need stricter provenance checks, permissions, and sandboxing than traditional app integrations.

Source: ESET Blog

CISA warned that attackers are actively exploiting Langflow flaws for credential harvesting, prompting the issue to be added to the Known Exploited Vulnerabilities (KEV) catalog. The update arrives shortly after Sysdig observed active targeting, underscoring how quickly agent-building platforms can turn into high-value compromise tools. Organizations using Langflow should prioritize patching and immediately review exposure, access controls, and credential handling around AI workflow deployments.

Source: Help Net Security

Wiz disclosed GhostApproval, a trust boundary weakness affecting multiple popular AI coding assistants that can allow booby-trapped repositories to take control of a developer’s machine. The core risk is that an assistant’s permission/approval flow can be manipulated so a “safe” change request results in writes to sensitive files. This is a reminder that agentic tooling requires strict security boundaries, safer file-write controls, and monitoring designed for autonomous behavior—not just “prompt” safety.

Source: Security Week

A threat actor posted claims on a cybercrime forum that it exfiltrated “just over 35GB” from Accenture, including source code, RSA/SSH keys, and multiple Azure access artifacts. Accenture has acknowledged the incident and stated it contained the impact, though the full extent remains unclear. For incident responders, this is a high-signal pattern: leaked source code plus API keys/tokens can rapidly accelerate follow-on attacks, so key rotation and access-token invalidation are urgent first steps.

Source: Help Net Security

Ubiquiti released updates addressing multiple critical UniFi OS vulnerabilities affecting the UniFi Connect application and other UniFi components. The most severe issue, CVE-2026-50746, is described as enabling command injection and privilege escalation, with versions affected listed by the vendor. Network operators should treat UniFi management interfaces as high-risk exposure points: patch quickly, restrict admin access, and review for indicators of exploitation.

Source: Security Affairs

Wireshark’s 4.6.7 maintenance release patches a dozen security flaws across multiple protocol dissectors and file-reading paths. Because packet analysis tools parse untrusted input at scale, they remain attractive targets for attackers seeking crashes or code-execution primitives. Teams running Wireshark on analyst workstations should update promptly and ensure capture files from untrusted sources are handled with least-privilege practices.

Source: Help Net Security

Zimperium’s zLabs describes RedWing as an Android spyware operation distributed and monetized through Telegram, including documentation and tutorial materials. The model shifts spyware from bespoke campaigns toward “service” distribution, potentially lowering barriers for new operators. The signal for defenders: monitor for unusual Android app distribution paths and strengthen mobile endpoint controls, even when the “delivery” looks like legitimate messaging infrastructure rather than classic malware hosting.

Source: Security Affairs

CISA urges immediate patching of exploited ColdFusion, Langflow, and Joomla flaws

Malicious AI agent skills can slip past scanners built to stop them

Chrome 150 update patches 27 vulnerabilities