This Month in Security, July 2026: A Record-Breaking Patch Tuesday, SharePoint Zero-Days Under Siege, and the First AI Agent Breach

July 2026 brought the largest Patch Tuesday in Microsoft's history, plus a run of actively exploited zero-days in enterprise infrastructure. The month is more likely to be remembered for something else: the first confirmed breach carried out by an autonomous AI agent, an attack defenders have been warning about for years. With AI now helping attackers find and exploit flaws quickly, the window between a patch and its exploit keeps getting shorter. Most notable with the Hugging Face Breach. The AI repo disclosed that attackers gained access to internal datasets and credentials after breaching their production infrastructure using an autonomous AI agent system (BleepingComputer).

‍

‍

Critical Zero-Day Exploits

‍

July saw emergency patches and active exploitation across SharePoint, network appliances, and email servers, with several flaws hitting CISA's Known Exploited Vulnerabilities catalog within days.

• SharePoint RCE (CVE-2026-50522): A critical deserialization flaw (CVSS 9.8) in on-premises SharePoint Server came under active exploitation after a public proof-of-concept was released. watchTowr observed attackers stealing IIS machine keys for persistent access, warning that patching alone is insufficient and that exposed credentials must be rotated. CISA added it to the KEV catalog on July 22 (The Hacker News)
• Check Point SmartConsole Bypass (CVE-2026-16232): Check Point patched an actively exploited authentication bypass in its SmartConsole admin panel that let unauthenticated attackers obtain an application login token and gain administrator privileges on internet-exposed Management Servers. CISA added the flaw to its KEV catalog and set a July 25 federal patch deadline (BleepingComputer)
•  SonicWall SMA1000 Chain (CVE-2026-15409 / CVE-2026-15410): An SSRF flaw chained with a command-injection bug was exploited as a zero-day as early as June 22 by a threat actor Volexity tracks as UTA0533, which deployed custom appliance malware dubbed KNUCKLEBALL, Sou5, and ORANGETAIL to gain root and covert access to SMA1000 VPN appliances (BleepingComputer)
• Windows LegacyHive Zero-Day: A Windows zero-day granting attackers administrative privileges received free, unofficial micropatches from 0patch after being observed in exploitation, giving defenders an interim fix ahead of an official Microsoft update (BleepingComputer)
• Zimbra Zero-Click Exploited by Russian Actors: CISA warned that the Russian state-sponsored group Laundry Bear (aka Void Blizzard) was combining phishing with exploitation of a now-patched Zimbra Collaboration zero-click flaw to steal email from targeted organizations (BleepingComputer)

‍

‍

Notable Threats and Incidents

‍

This month's incidents ran from AI-driven attacks to ransomware extortion, with victims across government agencies and critical infrastructure. The most notable event was the Hugging Face Breach.

• Hugging Face Breached by an Autonomous AI Agent: The AI platform disclosed that attackers breached its production infrastructure using an autonomous agent framework that executed thousands of actions across short-lived sandboxes, exploiting two code-execution flaws in its dataset pipeline to steal cloud credentials and. The agent stole cloud credentials and moved laterally, in what the company called the first real-world 'agentic attacker' scenario (BleepingComputer)
• Cl0p Targets PTC Windchill and FlexPLM (CVE-2026-12569): Cl0p ransomware affiliates chained a pre-authentication information disclosure with a server-side flaw to achieve unauthenticated RCE against internet-exposed PTC Windchill and FlexPLM PLM deployments, stealing engineering and design data from manufacturing, automotive, and aerospace targets for double extortion (The Hacker News)
• South Korea Diplomatic Academy Breach: South Korea disclosed that hackers spent ten months inside the National Diplomatic Academy's online education system, stealing personal information belonging to current and former Ministry of Foreign Affairs staff, including overseas diplomats (BleepingComputer)
• Origin Energy Data Breach: Australian energy provider Origin confirmed that an unauthorized party accessed and leaked customer data online, exposing sensitive personally identifiable information and underscoring the continued targeting of energy-sector providers (BleepingComputer)
• Stadler Rejects $12.3M Ransom: Swiss rail vehicle manufacturer Stadler said the Everest ransomware gang demanded roughly $12.3 million after breaching a data-exchange platform it shared with one of its suppliers. The breach came through a supplier rather than Stadler's own systems (BleepingComputer)
• Dolphin X Malware Ranks Victims with AI: A new Dolphin X remote access trojan advertised an AI-powered profiling feature that scores and ranks infected users, helping criminals prioritize the highest-value victims first. It is another case of AI being built straight into commodity malware (BleepingComputer)

‍

‍

Policy and Framework Updates

‍

Regulators and industry moved to address antitrust concerns and the security of the emerging AI agent ecosystem.

• EU Fines Google roughly $1 Billion Under the DMA: The European Commission fined Google 890 million euros (about $1 billion) after finding it violated the Digital Markets Act in its search and Play Store practices, continuing aggressive EU enforcement against gatekeeper platforms (BleepingComputer)
• NVIDIA-Led Open Secure AI Alliance Formed: NVIDIA and 36 other organizations, including Microsoft, Cisco, Cloudflare, CrowdStrike, IBM, Palo Alto Networks, Red Hat, and the Linux Foundation, launched the Open Secure AI Alliance to build shared, open technologies for securing AI agents across identity, permissions, isolation, guardrails, and logging (The Hacker News)
• GitHub Adds Dependabot Cooldown: GitHub introduced a three-day default cooldown before Dependabot opens version-update pull requests, aiming to limit the window in which a poisoned package release can be adopted downstream before it is pulled from the registry (security updates are still pushed immediately) (The Hacker News)

‍

‍

Patches and Vulnerabilities

‍

Microsoft Patch Tuesday: Fixed a record 622 CVEs, including two zero-days under active attack.

• Record 622 CVEs and Two Exploited Zero-Days: Microsoft's July release was its largest ever, more than tripling June's prior high. Two exploited privilege-escalation zero-days led the pack: CVE-2026-56164 in on-premises SharePoint Server (unauthenticated, network) and CVE-2026-56155 in Active Directory Federation Services, both credited to incident responders (The Hacker News)
• BitLocker Bypass and Kerberos RC4 Cleanup (CVE-2026-50661): The update also fixed a publicly disclosed BitLocker bypass requiring physical access and completed Microsoft's multi-year Kerberos RC4 hardening by removing the rollback switch. That change can break service-account logins unless admins audit and rotate RC4-reliant accounts first (The Hacker News)
• Fastjson 1.x RCE (CVE-2026-16723): Attackers began targeting a critical flaw (Alibaba-assigned CVSS 9.0) in the Fastjson Java library that allows unauthenticated code execution in affected Spring Boot applications. As of late July, no fixed Fastjson 1.x release was available, leaving mitigation as the only option (The Hacker News)
• vBulletin Pre-Auth RCE (CVE-2026-61511): Public exploit details revealed how an unauthenticated request can reach PHP's eval() function in vBulletin to execute code on unpatched forum servers. vBulletin had released fixes weeks earlier, making prompt upgrades to 6.2.2 essential (The Hacker News)

‍

‍

Key Takeaways for Staying Secure

‍

• Sort Patching by Exploitation, Not Severity: With Microsoft shipping 622 CVEs in a single month, CVSS scores no longer meaningfully triage risk. July's two exploited zero-days were mid-tier privilege flaws, not headline 9.8s. Prioritize using KEV, EPSS, and vendor 'exploited' flags, and patch faster than before.
• Rotate Credentials After SharePoint Patching: For CVE-2026-50522, patching does not evict attackers who already stole IIS machine keys. Rotate machine keys and any credentials on internet-facing SharePoint, and enable AMSI in Full Mode where supported.
• Prepare for Autonomous AI Attackers: The Hugging Face breach shows agentic attacks are no longer hypothetical. Monitor for high-volume automated activity from short-lived sandboxes, tighten pipeline input validation, and, as Hugging Face advises, keep a capable self-hosted model vetted and ready so guardrails don't block forensic work during an incident.
• Lock Down Exposed Management Interfaces: The Check Point and SonicWall zero-days both required internet-exposed management access. Restrict admin panels and appliance management to trusted IPs, and never expose them directly to the internet.
• Audit Kerberos RC4 Before Patching: The July update removes the RC4 rollback switch and can break authentication for service accounts still requesting RC4 tickets. Audit RC4 usage, rotate flagged service-account passwords to generate AES keys, then apply the update.
• Review Your Security Posture: When needed, reach out to a trusted provider to review your security posture, like Cloud Security Partners.

‍

‍

About the Author

‍

Jordan Darrah is a Security Consultant at Cloud Security Partners. Jordan's interest in IT started when she was working as a menswear fashion designer and bridal seamstress. Since then, she has built a diverse technical background spanning hardware repair, systems administration, regulatory compliance, and penetration testing.

Currently, Jordan specializes in application and cloud security assessments, where she evaluates system vulnerabilities and conducts penetration tests. Jordan holds multiple industry certifications, including CISSP, eJPT, and CompTIA PenTest+. She also runs an OSCP study group and maintains a blog where she breaks down concepts and tools for new security professionals.

‍