Microsoft Security Response Center

Attackers move fast. Getting protection in place can take longer than identifying the vulnerability. In his BlueHat Asia 2026 keynote, “Accelerated response by design,” Tom Gallagher (@secbughunter), VP of Engineering at MSRC, discusses that gap. AI and automation can help teams assess issues faster, but building and deploying a fix still takes time. Rapid mitigations can help reduce exposure in the meantime. Watch the keynote: msft.it/6015aCuMz View the slides: msft.it/6014aCuMK

Friday marked the final day of BlueHat Asia, with security professionals from across the community coming together to share research and exchange ideas. The day began with opening remarks from Jeff Moss (Founder, DEF CON & Black Hat), followed by a keynote from Tom Gallagher (VP of Engineering, MSRC), focused on practical security improvements, tactical mitigations, and raising the cost for attackers. From technical sessions to conversations in the Security Villages, Day 2 highlighted the value of collaboration in helping protect customers and strengthen security across the ecosystem. This video features highlights from the keynote by Tom Gallagher (@secbughunter), along with perspectives from Cameron Vincent (@SecretlyHidden1), Senior Security Researcher, MSRC, and Asem Eleraky @Melotover), Senior Penetration Tester, PayMob.

At BlueHat Asia, Chumy Tsai (@rm_rf_chumy) and Katherine Chen shared how they chained multiple SharePoint vulnerabilities into a full pre-authentication remote code execution (RCE) exploit. The research began with a JWT authentication bypass that enabled impersonation of arbitrary SharePoint users, including administrators, by abusing flaws in token validation and identity mapping. The second half of the talk focused on Flow2Shell (CVE-2026-47298), a SharePoint Workflow vulnerability. The researchers showed how workflows are built from XOML and rule files, then walked through four separate workflow validation mechanisms intended to prevent abuse. Through a series of bypasses involving type validation, rule validation, runtime checks, and workflow caching behavior, they were able to evade SharePoint's protections and execute malicious code on a fully patched server. One of the most interesting findings involved SharePoint's workflow cache. A malicious workflow assembly could be loaded into memory before validation errors were surfaced. By triggering the workflow a second time, SharePoint would reuse the cached assembly and skip portions of the validation pipeline, ultimately leading to code execution. Beyond the technical exploit chain, the talk highlighted the value of detailed vulnerability research. The complete proof-of-concept and root cause analysis enabled MSRC to quickly assess severity, identify additional variants, begin remediation, and better protect customers. As the presenters noted, impactful security research is not just about finding a bug, but proving impact, understanding root causes, and helping defenders address entire classes of vulnerabilities.

What happens when an AI assistant inherits the privileges of the user it's helping? At BlueHat Asia, Johann Rehberger (@wunderwuzzi23) discussed how SQL Copilot in SQL Server Management Studio can become a powerful target when connected to highly privileged accounts. Through his demos, he showed how "read-only" assumptions can break down, how prompt injection techniques can influence AI behavior, and why security boundaries matter just as much as model instructions. The talk offered a deep dive into AI security, database permissions, and the unintended risks that emerge when assistants are granted access to sensitive environments. Johann Rehberger's key message was simple: trust, permissions, and execution context matter.

How do cross-tenant vulnerabilities happen in complex cloud environments? At BlueHat Asia, Haakon Wik Gulbrandsrud walked through his research into Azure Logic Apps and API connections, showing how a single architectural root cause led to multiple cross-tenant security issues. Through a series of real-world examples, he demonstrated how understanding platform internals, not just hunting for bugs, can uncover high-impact vulnerabilities hiding in plain sight. The talk offered a fascinating look at the research process itself: tracing connections, mapping architecture, and asking what could go wrong at every step until the answers lead somewhere unexpected.

At BlueHat Asia, Salim Chawro, Corporate Vice President of Microsoft Cloud Security, and Zeeshan Syed, Vice President of Microsoft Cloud Security, shared how Microsoft is rethinking proactive defense as cloud environments continue to grow in size and complexity. They argued that traditional point-in-time security reviews can no longer keep pace with services that are constantly changing and threat landscapes that are accelerating with AI. The talk introduced Argus, Microsoft's agentic security assurance system, which brings together code, identities, configurations, dependencies, runtime signals, and architectural context to help teams identify control gaps, uncover root causes, model attack paths, and prioritize remediation. Rather than focusing on individual vulnerabilities, the approach helps teams understand how risks connect across an entire service and address broader classes of security issues. The speakers also discussed how defenders have a unique advantage in the AI era: access to the full context of their environments. By combining that context with agentic workflows, Microsoft is working to deliver continuous security assurance at a scale that would be impossible through manual review alone. With more than 90% of Argus findings confirmed as genuine security issues, the system is helping teams conduct deeper reviews in hours instead of weeks.