July 24, 2026 is shaping up as one of the more consequential single-day threat briefings of the year. Six critical disclosures span nation-state phishing, a publicly weaponized Active Directory privilege escalation, a rogue AI agent deployment flaw, a systemic server-side image processing vulnerability, autonomous AI-assisted intrusion, and active ransomware targeting industrial software. The common thread running through all of them is speed: attackers are compressing the time between initial access and full domain or data compromise, and in at least one case they have removed the human from the loop entirely.

Starting with identity and Active Directory, the Certighost exploit is the most structurally dangerous item on today's list. Any low-privileged domain user can now use a publicly available tool to obtain a Domain Controller certificate through misconfigured Active Directory Certificate Services templates, then authenticate as that DC and execute a DCSync attack to pull the krbtgt hash. Once an attacker holds the krbtgt hash, they can forge Kerberos Golden Tickets and maintain persistent, nearly invisible access to the entire domain. This is not a theoretical risk — the exploit is public. Security teams should treat this as a fire drill. Audit every ADCS certificate template immediately, focusing on templates that allow requestor-supplied subject names combined with client authentication EKUs. Microsoft's PKI Health Tool and tools like Certify or PSPKIAudit can surface vulnerable templates. Disable or restrict enrollment on any template that does not require CA manager approval until it has been reviewed. Monitor for unusual certificate enrollment events in the Certificate Authority logs and alert on any DC authentication originating from a non-DC machine account.

The BlueNoroff Zoom and Teams phishing campaign is a sharp reminder that North Korean APT groups have not slowed their cryptocurrency targeting. The kit uses typosquatted domains mimicking Zoom and Microsoft Teams, combined with ClickFix-style social engineering that tricks users into running commands themselves — bypassing many endpoint controls. The campaign specifically profiles crypto wallet holders before delivering malware, suggesting reconnaissance-driven targeting rather than spray-and-pray. Defenders should block newly registered domains at the DNS and proxy layer, enforce application allowlisting so that users cannot execute arbitrary scripts from browser prompts, and train staff to recognize ClickFix lures, which typically present a fake CAPTCHA or error screen with copy-paste instructions. Cryptocurrency custodians and finance teams are the highest-priority groups for awareness reinforcement.

The AgentForger flaw in ChatGPT Workspace Agents, now patched by OpenAI as of June 8, deserves continued attention because many organizations have not audited what happened in the window before patching. A single phishing link could silently create and authorize rogue AI agents inside a victim's workspace. Teams should review their ChatGPT Workspace agent inventory, revoke any agents created by unexpected principals, and audit OAuth token grants made during the exposure window. Going forward, enforce a change-approval workflow for any new agent deployment and log all agent creation events centrally.

The Bing image processing vulnerability — where a crafted SVG triggered SYSTEM-level command execution across multiple production Windows and Linux hosts — illustrates the risk of trusting user-supplied file content in server-side pipelines. While this was a Microsoft infrastructure issue rather than a customer-deployed product, it reinforces a critical principle: any service that processes user-uploaded files must do so in isolated, least-privilege sandboxes with no network egress to internal systems. Organizations running their own image or document processing pipelines should review sandbox isolation, enforce strict content-type validation, and strip active content from SVG files before processing.

The Hermes autonomous AI agent attack against Thailand's Ministry of Finance is a landmark event. A threat actor deployed an unattended AI agent that performed privilege escalation, host enumeration, and file traversal without requiring human approval at each step. This compresses attacker dwell time dramatically and means that traditional detection windows are shrinking. Defenders should focus on behavioral analytics that flag rapid, sequential enumeration activity — even if each individual action looks benign — and ensure that privileged command execution triggers immediate alerting rather than batch review.

Finally, Clop ransomware is actively scanning for and exploiting internet-exposed PTC Windchill and FlexPLM instances to steal product lifecycle and manufacturing data. If your organization runs either platform, check external exposure immediately.

Defensive priorities for today:
- Audit and restrict ADCS certificate templates; treat Certighost as an active threat requiring same-day response
- Block typosquatted collaboration-tool domains at DNS and proxy; brief crypto and finance staff on ClickFix lures
- Audit ChatGPT Workspace agent inventory and revoke unexpected grants from the AgentForger exposure window
- Review server-side file processing pipelines for sandbox isolation and SVG content stripping
- Deploy behavioral detection rules for rapid sequential enumeration to counter autonomous AI-assisted intrusion
- Immediately remove PTC Windchill and FlexPLM from public internet exposure or place behind authenticated VPN with WAF inspection

This briefing is informational and does not replace official vendor advisories, patches, or guidance from your organization's security leadership.