July 24, 2026 marks a convergence of threats that security teams cannot afford to treat in isolation. Today's critical items share a common thread: attackers are systematically dismantling trust — trust in collaboration tools, trust in identity infrastructure, trust in AI platforms, and trust in internet-facing enterprise software. Each of these vectors is being exploited with increasing sophistication, and in at least one confirmed case, without any human attacker sitting at a keyboard.

The most operationally urgent item for enterprise defenders is Certighost, a publicly available exploit targeting misconfigured Active Directory Certificate Services. Any low-privileged domain user can leverage this to obtain a Domain Controller certificate, authenticate as that DC, and execute a DCSync attack to extract the krbtgt hash — effectively achieving full domain compromise. The word 'publicly available' cannot be overstated here. This is not a theoretical research finding; it is a weaponized tool that any moderately skilled attacker can run today. Security teams should immediately audit their AD CS configuration using tools like Certify or PSPKIAudit to identify misconfigured certificate templates, particularly those with the ENROLLEE_SUPPLIES_SUBJECT flag enabled or overly permissive enrollment rights. Prioritize patching and template hardening before any other item on this list.

Running in parallel, BlueNoroff — the North Korean APT group with a well-documented focus on financial theft — is operating a phishing kit that impersonates Zoom and Microsoft Teams through typosquatted domains. The campaign uses ClickFix-style social engineering, a technique that tricks users into manually executing malicious commands under the guise of fixing a technical problem. The targeting is precise: cryptocurrency wallet holders and individuals in the digital asset industry. Defenders should push DNS filtering rules for known typosquatted collaboration domains, enforce application allowlisting to block unexpected script execution, and brief high-risk users — particularly anyone in finance or crypto-adjacent roles — on ClickFix-style lures. Endpoint detection rules that flag PowerShell or cmd.exe spawned from browser or meeting-client processes are highly relevant here.

The ChatGPT AgentForger vulnerability, now patched by OpenAI as of June 8, deserves continued attention because patch confirmation does not equal organizational safety. The flaw allowed a single phishing link to silently create and authorize rogue AI agents within a victim's ChatGPT Workspace environment. Organizations should audit their current AI agent inventory, review which agents have been granted permissions to external tools or data sources, and implement approval workflows requiring human review before any new agent is deployed or authorized. This incident is a preview of an emerging attack class: AI supply chain compromise through agent injection.

The Bing image processing vulnerability — where a crafted SVG file triggered arbitrary command execution as SYSTEM on Windows workers and root on Linux workers across multiple production hosts — illustrates the systemic risk of processing untrusted file formats at scale. While Microsoft has addressed this, the lesson for any organization running image or document processing pipelines is clear: sandbox all file parsing in isolated, least-privilege environments, enforce strict content-type validation before processing, and monitor for unexpected process spawning from media-handling services.

Perhaps the most strategically significant item today is the confirmed deployment of an autonomous AI agent called Hermes against Thailand's Ministry of Finance. The attacker ran this agent in unattended post-exploitation mode, enabling automated privilege escalation, host enumeration, and file system traversal without any human approval gate. This is the first widely reported case of AI-driven post-exploitation operating at this level of autonomy against a government target. Defenders must now account for the speed differential: AI-assisted attackers can compress the time between initial access and data exfiltration dramatically. Behavioral detection rules that flag rapid sequential enumeration commands, unusual LDAP queries, and bulk file access patterns are essential countermeasures.

Finally, Clop ransomware is actively targeting internet-exposed instances of PTC Windchill and FlexPLM — platforms used for product lifecycle management and manufacturing operations. Clop's focus is data exfiltration for extortion rather than encryption alone, meaning even organizations with strong backup postures are at risk. Any Windchill or FlexPLM instance with public-facing exposure should be placed behind a VPN or zero-trust access gateway immediately, with emergency patching prioritized.

Defensive priorities for today:

- Audit and harden AD CS certificate templates immediately; treat Certighost as an active exploitation risk
- Deploy DNS and proxy blocks for typosquatted Zoom and Teams domains; brief crypto-adjacent staff on ClickFix lures
- Audit ChatGPT Workspace agent permissions and enforce human-in-the-loop approval for all agent deployments
- Sandbox all untrusted file processing pipelines and monitor for anomalous process execution from media handlers
- Implement behavioral detections for rapid enumeration and bulk file access consistent with AI-assisted post-exploitation
- Remove PTC Windchill and FlexPLM from direct internet exposure and apply all available vendor patches without delay

This briefing is informational and does not substitute for official vendor advisories, CVE disclosures, or guidance from your organization's incident response team.