AI Security Platforms Rise as DDoS, Zero‑Days and PoCs Intensify
Coverage: 27 Jul 2026 (UTC)
< view all daily briefs >Major cloud and platform vendors advanced AI-driven security, data access, and resilience, while defenders faced active exploitation and new proofs-of-concept across widely used software. Microsoft outlined an agentic approach to risk reduction, NVIDIA rallied partners around open-secure AI, and AWS introduced both workflow and DDoS protection changes. At the same time, urgent patches and mitigations landed for network orchestration, Java deserialization, web forums, and automation platforms.
AI-Driven Defense and Open Ecosystems
Microsoft proposed a new Cyber Stack centered on agentic automation with Microsoft blog detailing Project Perception, a system that coordinates red (offensive path discovery), blue (investigation and risk evaluation), and green (remediation and hardening) agents. The approach relies on broad visibility across identities, endpoints, applications, data, clouds, and AI systems, and uses a multi-model architecture to select models by quality, latency, reliability, and cost. Its first public scenario targets software vulnerability management using MAI-Cyber-1-Flash within MDASH, with public preview beginning August 3 under Microsoft’s Responsible AI and enterprise governance controls.
NVIDIA and 36 partners formed the Open Secure AI Alliance, with The Hacker News reporting the coalition aims to develop and share open technologies for securing AI agents and their software stack. NVIDIA’s initial contribution is the open-source NOOA framework for testing, tracing, auditing, and governing agent behaviors. The announcement links urgency to the July Hugging Face intrusion, highlighting the operational advantage of locally controlled models for incident response. Public materials do not yet describe the alliance’s charter, governance, joint roadmap, or shared repositories.
AWS introduced the preview of the AWS Security Hub MCP App via a local Model Context Protocol server that feeds Security Hub exposure findings into Claude Desktop, with AWS update noting read-only operation on the user’s machine, use of existing AWS credentials, and no additional cost. Analysts can query exposures in natural language, inspect attack paths and resource configurations, and receive remediation recommendations, combining concise AI-ready summaries with interactive visualizations in the same conversation.
Data and Cloud Foundations for AI
SAP and Google Cloud announced general availability of SAP Business Data Cloud Connect for BigQuery, delivering zero-copy, bi-directional access between SAP data and Google’s data and AI ecosystem. The Google Cloud blog describes live querying of SAP tables with preserved semantics directly in BigQuery and Knowledge Catalog, reduced data duplication and costs, and support for agentic AI workflows that act on real-time ERP signals. The capability is available globally for Google Cloud customers, with SAP Business Data Cloud region availability and cross-cloud support for instances on Google Cloud and AWS (Azure support planned).
Amazon RDS for SQL Server removed a long-standing restore limitation for encrypted databases. An AWS update confirms customers can now restore Transparent Data Encryption (TDE)-enabled databases to Multi-AZ instances and to instances with a same-region read replica using native backup/restore. By backing up TDE certificates to Amazon S3 and restoring them alongside the encrypted backups, organizations can streamline migrations and recovery while maintaining encryption at rest and high availability across all AWS Regions that support Amazon RDS for SQL Server.
DDoS Defense Shifts and Botnets
AWS is migrating Shield Advanced’s application-layer automatic mitigation to the AWS blog announced AWS WAF Anti-DDoS managed rule group. Eligible web ACLs enter Count mode from July 27–August 7, 2026, followed by a free evaluation through September 30; auto-upgrades start October 1, with guided migration through December 31. Legacy application-layer automatic mitigation retires January 1, 2027. The rule group adds a Challenge action (including silent browser challenges), independent sensitivity controls for Block and Challenge, URI regex exemptions, and richer observability via metrics and request labels. It also reduces WCU consumption from 150 to 50 and excludes blocked DDoS requests from WAF, Anti-DDoS, and Shield Advanced billing during active mitigation. AWS notes the migration runs alongside existing protections in Count mode to validate parity before switching.
Separately, researchers tracked a rapidly evolving DDoS and proxy botnet dubbed Dysphoria that has infected an estimated 200,000 devices globally. According to BleepingComputer, the malware leverages blockchain-based domain resolution via ENS and SNS, obfuscates C2 data in IPv6-like strings, and spreads through weak Telnet/SSH credentials and a mix of recent and older device vulnerabilities. Recent variants focus on converting devices into network proxies by abusing UPnP, and operators claim up to 4 Tbps DDoS capacity. Recommended mitigations include firmware updates, changing default passwords, disabling unnecessary remote access, and hardening device configurations.
Critical Vulnerabilities Actively Targeted
Arista patched a critical unauthenticated OS command injection in on-premises VeloCloud Orchestrator (VCO), tracked as CVE-2026-16812 (CVSS 10.0), with BleepingComputer reporting active exploitation. Fixed versions include 5.2.3.14, 6.1.3.4, 6.4.2.4, and 7.0.0.1 and later; affected builds span 5.2.x before 5.2.3.14, 6.1.x before 6.1.3.4, 6.4.x before 6.4.2.4, and early 7.0.x. Hosted and dedicated VCO deployments and VeloCloud Gateway/Edge are not vulnerable. CISA added the flaw to its KEV catalog and directed federal agencies to mitigate by July 30, 2026. Administrators are advised to patch immediately, restrict VCO web access to admin networks, preserve logs, rotate credentials, validate or restore managed devices, and examine traffic for unusual requests or command execution, including activity from 8.19.75.217, 206.72.242.124, and 206.72.242.162.
Researchers disclosed a proof-of-concept for "Certighost" (CVE-2026-54121), an Active Directory Certificate Services flaw that can let low-privilege users impersonate a domain controller by abusing a fallback "chase" mechanism during enrollment. As detailed by BleepingComputer, a rogue host can respond to CA lookups using attacker-specified cdc and rmd attributes to return forged identity data, leading the CA to issue a certificate usable with PKINIT. Microsoft addressed the issue in July 2026 by validating that the cdc host maps to a legitimate domain controller and that identity attributes match expectations. A temporary mitigation disables the chase fallback via certutil, though researchers caution it is untested in production.
Active exploitation is also underway against FastJson (CVE-2026-16723), a critical remote code execution flaw affecting versions 1.2.68 through 1.2.83. BleepingComputer reports attackers are achieving code execution in common Spring Boot fat-JAR deployments by abusing @type processing before AutoType restrictions apply, without requiring third-party gadget chains or user interaction. With FastJson 1.x no longer actively maintained and no vendor patch available, users should enable SafeMode, migrate to fastjson2 (allowlist-first polymorphic deserialization), or change deployment patterns to reduce exposure. Campaigns reportedly focus on U.S. organizations across multiple sectors, with some incidents in Singapore and Canada.
Separately, public exploit details landed for vBulletin’s template-engine pre-auth RCE, CVE-2026-61511. The The Hacker News coverage notes the bug in vB5_Template_Runtime::runMaths() filters expressions and then passes them to PHP’s eval(), allowing attackers to reconstruct function names without letters and reach system() via ajax/render/pagenav. vBulletin released fixes and version 6.2.2 on July 1; cloud-hosted instances are reported patched, leaving self-hosted forums at risk if unupdated. Defenders should update promptly and monitor for unusually long, operator-heavy pagenav[pagenumber] POST values as an indicator.
In the automation space, Security Joes detailed a high-severity expression-sandbox escape in n8n (GHSA-gv7g-jm28-cr3m, CVSS 8.7). According to The Hacker News, an authenticated workflow editor could execute OS commands as the n8n process by exploiting gaps in the expression rewriter (concise arrow functions and Reflect.get() property access), reaching Node.js internals to load child_process. Fixes shipped in 2.31.5 and 2.32.1; administrators should apply patches, review recently created or modified workflows for suspicious constructs, hunt for spawned shells or network tools as children of n8n/Node.js, and rotate credentials if anomalies appear.