Puppet RCE, Java4Shell, and Optimizely Malware; Microsoft's Cheap
This digest covers critical RCE vulnerabilities in Puppet and Java, a widespread malware campaign via Optimizely, and Microsoft's new cost-effective AI coding
Today's news highlights a tension between the increasing sophistication of AI-driven tools and persistent, critical vulnerabilities in widely used infrastructure. The most significant stories involve severe RCE flaws in Puppet and Java, requiring immediate patching, alongside a malware incident leveraging Optimizely's platform.
TL;DR
- Hackers Actively Exploiting Critical RCE Flaw in Perforce Puppet Enterprise and OpenVox — A critical deserialization RCE vulnerability (CVE-2026-115731) in Perforce Puppet Enterprise and OpenVox is being actively exploited, prompting a CISA alert.
- FBI and AWS Warn of AI Agents Automating Cybercrime at 'Industrial Scale' via ARTEX Framework — FBI and AWS warn that attackers are using the open-source ARTEX AI penetration testing framework to automate cybercrime, including stealing AI credentials and generating deepfake videos for KYC bypass.
- CISA Recommends Immediate Patching for Java4Shell Vulnerability Allowing RCE in Oracle Java — CISA urges immediate patching for CVE-2026-114938, known as Java4Shell, a critical out-of-bounds write flaw in Oracle Java's Base64 decoder that can lead to remote code execution.
- Optimizely Used for A/B Testing Malware on 178 Websites Including BBC, CNN, and Zoom — Attackers exploited a vulnerability in Optimizely Experimentation to inject malicious JavaScript, serving malware to visitors of 178 websites, including BBC, CNN, and Zoom.
- Microsoft Introduces New AI Coding Model Claiming Near Claude Opus 5.5 Performance at 5% Cost — Microsoft unveiled MAI Code 2 and MAI Code 2 Flash models, claiming MAI Code 2 achieves 96% of Claude Opus 5.5's performance for 5% of its cost in coding tasks.
- FBI Reports ARTEX AI Hacking Tool Used to Breach Networks in Under Three Hours — The FBI warns that the open-source ARTEX AI hacking tool has been used by attackers to breach networks in under three hours.
- Microsoft's New Coding AI Model Targets High Performance at Reduced Cost — Microsoft's new coding model aims to deliver near-Opus performance for developer tasks at a significantly lower cost.
- Proton Launches Privacy-Focused Lumo AI Desktop App for Local Task Automation — Proton launched Lumo Desktop for Windows, macOS, and Linux, offering a privacy-focused AI agent for local tasks with user-controlled access to files and apps.
- Microsoft 365 Suffers Nearly 24-Hour Outage Due to Combination of Three Failures — A latent cloud gateway bug, traffic shifts, and reduced server capacity combined to cause a near 24-hour Microsoft 365 outage affecting Exchange infrastructure.
- Vercel Releases UI SDK for AI Agents to Programmatically Compose User Interfaces — Vercel has released a UI SDK for AI agents, allowing them to programmatically compose user interfaces from reusable UI behaviors.
Hackers Actively Exploiting Critical RCE Flaw in Perforce Puppet Enterprise and OpenVox
Top story · Security · The Hacker News · 2026-11-18

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added CVE-2026-115731, a critical security flaw in Perforce Puppet, to its Known Exploited Vulnerabilities catalog due to active exploitation. The vulnerability is a deserialization of untrusted data issue that allows remote code execution on the Puppet Server.
An attacker with any CA-signed agent certificate can exploit this by submitting a crafted rich-data type through a PUT request. If certificate authorities automatically sign requests without strong verification, unauthenticated network attackers can obtain valid certificates and exploit the flaw. Patches were released by Perforce Software on November 3, 2026, and the issue also affects OpenVox.
The flaw impacts Puppet Enterprise versions before 2026.0.1, 2025.12.1, and 2023.8.13, and Puppet Core versions before 9.1.3 and 8.22.2. A related certificate-name authorization bypass (CVE-2026-115732) also affects both Puppet and OpenVox.
Key facts
- CVE-2026-115731 CVSS score: 9.4
- Patched by Perforce Software on November 3, 2026
- Affects Puppet Enterprise versions before 2026.0.1, 2025.12.1, 2023.8.13
- Affects Puppet Core versions before 9.1.3, 8.22.2
- Affects OpenVox Agent 9.3.0, 8.32.0 and OpenVox Server 9.2.0, 8.18.0
Why it matters: Engineers using Puppet or OpenVox for infrastructure-as-code and configuration management must immediately patch their systems to prevent unauthenticated remote code execution on Puppet servers, especially if certificate authority automatically signs requests.
Actively exploited RCE in Puppet and OpenVox requires immediate patching of affected versions to prevent server compromise.
🔗 Read more at The Hacker News
FBI and AWS Warn of AI Agents Automating Cybercrime at 'Industrial Scale' via ARTEX Framework
Security · The Register · 2026-11-17

The FBI and Amazon Threat Intelligence have issued alerts stating that attackers are using ARTEX, an open-source AI penetration testing framework, to automate cyberattacks against critical sectors. Between October 3 and November 5, Amazon's MadPot honeypot detected two ARTEX campaigns.
One campaign used AI agents to steal AI API keys and cloud credentials to gain free access to models, targeting LLM gateways and proxy platforms like One API, New API, and LiteLLM, and AI application builders such as Dify and FastGPT. The second campaign involved using ARTEX to create real-time deepfake videos to bypass identity verification (KYC) checks at two cryptocurrency exchanges and forge ID photos for document checks. Although the deepfakes were rejected, the agents achieved 11.5 frames per second on a consumer-grade GPU.
Key facts
- ARTEX used in campaigns between October 3 and November 5
- First campaign stole AI API keys and cloud credentials
- Second campaign generated real-time deepfakes for KYC bypass
- Deepfake video achieved 11.5 frames per second on consumer-grade GPU
- Targets included LLM gateways (e.g., One API, LiteLLM) and AI app builders (e.g., Dify, FastGPT)
Why it matters: Security and AI engineers must be aware of ARTEX and similar AI-driven frameworks as they enable automated, industrial-scale cybercrime. This necessitates stronger defenses for API keys, AI model access, and enhanced biometric verification to counter sophisticated deepfake attacks.
AI agents, powered by frameworks like ARTEX, are enabling cybercriminals to automate attacks at an industrial scale, demanding enhanced security measures for AI systems and identity verification.
CISA Recommends Immediate Patching for Java4Shell Vulnerability Allowing RCE in Oracle Java
Security · The Hacker News · 2026-11-18

CISA has advised organizations to immediately patch CVE-2026-114938, a critical out-of-bounds write vulnerability in Oracle Java, dubbed 'Java4Shell.' This flaw, CVSS score 8.1, affects the Base64 decoder of the x86-64 HotSpot Java Virtual Machine (JVM).
Specially crafted Base64 input can crash the JVM and enable unauthenticated remote code execution. Patches have been released by Oracle and OpenJDK. While no active exploitation has been confirmed, Java4Shell can allow attackers to execute arbitrary code on platforms like Keycloak, Atlassian Jira, Confluence, JFrog Artifactory, Jenkins, and PingFederate without authentication under default settings.
Key facts
- CVE-2026-114938 CVSS score: 8.1
- Affects Base64 decoder of x86-64 HotSpot Java Virtual Machine (JVM) in JDK
- Patches released by Oracle and OpenJDK
- Can enable unauthenticated RCE on Keycloak, Jira, Confluence, Artifactory, Jenkins, PingFederate
Why it matters: Java engineers and system administrators must prioritize patching Oracle Java installations immediately to prevent potential denial-of-service or unauthenticated remote code execution, particularly on critical enterprise applications.
Patch Java now: the critical Java4Shell vulnerability in the Base64 decoder can lead to unauthenticated RCE.
🔗 Read more at The Hacker News
Optimizely Used for A/B Testing Malware on 178 Websites Including BBC, CNN, and Zoom
Security · Krebs on Security · 2026-11-18

Over the past 24 hours, at least 178 prominent websites unknowingly served malware to visitors due to a security vulnerability in Optimizely, a company providing A/B testing and user experience customization. The incident, observed on November 16, involved malicious JavaScript code being delivered through Optimizely's platform.
This code injected data-stealing malware, including kits that could steal payment card data, cookies, and screenshots, and swap cryptocurrency addresses. Custom exploit code targeting known vulnerabilities in older Apple iPhone and iPad devices was also served. Optimizely confirmed a security incident on November 16, stating an unauthorized party exploited a vulnerability in Optimizely Experimentation and that the issue has been fully remediated.
Key facts
- Incident occurred on November 16, 2026
- 178 websites affected, including BBC, CNN, Equifax, Ledger, Zoom
- Malware payload rebuilt 26 times during the day
- Malware stole payment card data, cookies, screenshots, swapped crypto addresses
- Exploited vulnerabilities in older Apple iPhone/iPad devices
Why it matters: Backend and systems engineers need to enforce stricter Content Security Policies (CSPs) on third-party JavaScript to mitigate supply chain attacks, as even trusted vendors like Optimizely can become vectors for malware delivery.
Third-party JavaScript from trusted vendors like Optimizely can become a critical attack vector, underscoring the need for strict content security policies.
🔗 Read more at Krebs on Security
Microsoft Introduces New AI Coding Model Claiming Near Claude Opus 5.5 Performance at 5% Cost
AI Models · ZDNET · 2026-11-18

At Ignite 2026, Microsoft announced two new frontier AI models, MAI Code 2 and MAI Code 2 Flash, designed for "everyday" coding tasks. Microsoft claims that its MAI Code 2 model performs at 96% of Claude Opus 5.5's level while costing only 5% of Opus 5.5's price.
The MAI Code 2 Flash model is said to be comparable to lower-end models like GPT-6 Luna and Claude Haiku 5.5, also at a considerably lower cost. These are usage-based metrics for API usage. The benchmarks are provided by Microsoft.
Key facts
- MAI Code 2 announced at Ignite 2026
- Claims 96% of Claude Opus 5.5 performance
- Claims 5% of Claude Opus 5.5 cost
- MAI Code 2 Flash comparable to GPT-6 Luna and Claude Haiku 5.5
Why it matters: AI and backend engineers evaluating coding AI models should investigate Microsoft's new MAI Code 2 for potential significant cost savings without a drastic performance drop, which could impact development budgets and model choices.
Microsoft's new MAI Code 2 promises near-top-tier coding AI performance at a fraction of the cost, potentially disrupting the market for AI development tools.
FBI Reports ARTEX AI Hacking Tool Used to Breach Networks in Under Three Hours
Security · TechCrunch · 2026-11-16

The FBI has reported that the open-source AI penetration testing tool, ARTEX, is being used by attackers to breach networks. The agency noted that these breaches have occurred in under three hours, indicating a rapid and automated attack capability.
This highlights the increasing speed and efficiency with which AI-powered tools can be leveraged in cyberattacks, posing a significant threat to network security.
Key facts
- ARTEX used to breach networks in under three hours
Why it matters: This rapid compromise capability means security engineers need to accelerate their threat detection and response times, leveraging automation and real-time monitoring to counter AI-driven attacks that can breach systems before manual intervention is possible.
AI-powered hacking tools like ARTEX can breach networks in hours, demanding faster, automated defensive strategies.
Microsoft's New Coding AI Model Targets High Performance at Reduced Cost
Developer Tools · The New Stack · 2026-11-18

Microsoft has introduced a new coding model intended to provide high performance for developers while drastically cutting costs. The company's goal is to offer an efficient AI solution for everyday programming needs.
This initiative could impact the broader market for AI-powered developer tools by making advanced capabilities more accessible from a cost perspective.
Why it matters: This development suggests future AI-assisted development tools could become more affordable and widely adopted, prompting engineers to re-evaluate their current LLM toolchains and budget allocations for AI services.
Microsoft is aiming to make high-performance AI coding assistance much more affordable.
Proton Launches Privacy-Focused Lumo AI Desktop App for Local Task Automation
AI Agents & Tooling · TechCrunch · 2026-11-17

Proton, known for its privacy-focused alternatives to major tech services, has released Lumo Desktop for Windows, macOS, and Linux. This AI app allows users to automate local tasks, interact with connected applications, and manage personal files.
Lumo is designed to give users granular control over their data, emphasizing privacy by allowing choices on which actions the AI agent takes requiring approval. Proton states that conversations and connected app/file information are not used for personalized ads or AI model training. The app includes connectors for file-system access, shell execution, browser content fetching, Proton Mail, Confluence, Dropbox, and Jira. Proton uses encryption for chat history, stores no server-side logs, and hosts data in Europe under Swiss privacy law. Proton is also investing in infrastructure, with its first NVIDIA B300 cluster built on Open Compute Project (OCP) standards now live in Switzerland.
Key facts
- Lumo Desktop available for Windows, macOS, Linux
- Includes connectors for file-system, shell, browser, Proton Mail, Confluence, Dropbox, Jira
- Supports local and remote MCP (Model Context Protocol) servers
- First NVIDIA B300 cluster on OCP standards in Switzerland now live
- Data hosted in Europe under Swiss privacy law, outside U.S. CLOUD Act
Why it matters: Engineers concerned with data privacy in AI agent deployments should evaluate Proton's Lumo Desktop as a local, user-controlled option, especially given its Swiss data hosting and no-logging policy. Its connectors offer integration points for existing systems.
Proton's Lumo brings privacy-focused AI agents to the desktop, offering granular control over local data access and task automation.
Microsoft 365 Suffers Nearly 24-Hour Outage Due to Combination of Three Failures
Cloud & Infrastructure · Ars Technica · 2026-11-17

Microsoft 365 and Exchange infrastructure experienced a nearly 24-hour outage from November 12 to November 13, 2026. Microsoft's preliminary Post-Incident Report (PIR) attributes the outage to a combination of three factors: a hidden defect in its cloud gateway, standard network maintenance traffic shifts, and reduced server capacity.
The defect, present since July, caused CPU and memory exhaustion under certain conditions, exacerbated by higher CPU core counts in production systems. Traffic from Teams was shifted to the primary gateway 15 hours prior, exposing the bug and increasing load with a different request mix. An existing fault meant the gateway server fleet was operating at 14% below optimal, leading to lock contention and over 80% of affected gateway servers freezing.
Key facts
- Outage from November 12 at 16:20 UTC to November 13 at 16:10 UTC (just under 24 hours)
- Latent bug in cloud gateway deployed in July
- Gateway server fleet operating 14% below optimal capacity
- Incident IDs: EX1550543, MO1550821
Why it matters: This incident highlights the importance of comprehensive testing environments that mirror production, robust capacity planning, and diversified failover strategies for backend and cloud engineers managing critical distributed systems, even for major cloud providers.
Even small, hidden defects can combine with routine operations and capacity issues to trigger widespread outages in complex cloud infrastructure.
Quick hits
- Vercel Releases UI SDK for AI Agents to Programmatically Compose User Interfaces (InfoQ) — Vercel has released a UI SDK for AI agents, allowing them to programmatically compose user interfaces from reusable UI behaviors.
What to watch
- The industry's independent validation of Microsoft's MAI Code 2 performance and cost claims will be crucial to understand its true market impact and potential for wider adoption.
- How organizations implement stricter Content Security Policies (CSPs) and other third-party script protections in response to the Optimizely incident to mitigate supply chain attacks.
- The evolution of defenses against AI-driven cybercrime frameworks like ARTEX, particularly in hardening API and cloud credential access, and improving liveness detection in identity verification.
Sources
- Hackers Actively Exploiting Critical RCE Flaw in Perforce Puppet Enterprise and OpenVox — The Hacker News, 2026-11-18
- FBI and AWS Warn of AI Agents Automating Cybercrime at 'Industrial Scale' via ARTEX Framework — The Register, 2026-11-17
- CISA Recommends Immediate Patching for Java4Shell Vulnerability Allowing RCE in Oracle Java — The Hacker News, 2026-11-18
- Optimizely Used for A/B Testing Malware on 178 Websites Including BBC, CNN, and Zoom — Krebs on Security, 2026-11-18
- Microsoft Introduces New AI Coding Model Claiming Near Claude Opus 5.5 Performance at 5% Cost — ZDNET, 2026-11-18
- FBI Reports ARTEX AI Hacking Tool Used to Breach Networks in Under Three Hours — TechCrunch, 2026-11-16
- Microsoft's New Coding AI Model Targets High Performance at Reduced Cost — The New Stack, 2026-11-18
- Proton Launches Privacy-Focused Lumo AI Desktop App for Local Task Automation — TechCrunch, 2026-11-17
- Microsoft 365 Suffers Nearly 24-Hour Outage Due to Combination of Three Failures — Ars Technica, 2026-11-17
- Vercel Releases UI SDK for AI Agents to Programmatically Compose User Interfaces — InfoQ, 2026-11-18