Fireworks AI has confirmed a significant security breach involving its infrastructure, resulting in the unauthorized access of sensitive information. The incident impacted customer datasets, model weights, and internal Notion AI chat logs. This disclosure follows reports of a broader compromise involving the SGLang framework, which was exploited to execute remote code on the NEAR AI platform. That specific attack resulted in the theft of data from approximately 100 users and 18 distinct AI agents.
The security landscape for AI-driven platforms has faced mounting pressure as researchers identify new vectors for exploitation. In a separate development, the agentic pentesting platform known as AntiFox has been observed being weaponized by malicious actors. Since November 10, 2026, threat actors have utilized the tool to exfiltrate Amazon Web Services (AWS) credentials and perform lateral movement within compromised network environments.
These incidents highlight a growing trend of AI tools being repurposed for offensive operations. While the industry continues to explore the benefits of autonomous agents, the risk of these systems being turned against their owners is increasing. Reports indicate that open-weight AI models are becoming more proficient at executing cyberattacks, raising concerns about the security of the development lifecycle for these technologies.
The vulnerability of AI infrastructure is further underscored by recent research into cryptographic security. Google DeepMind has demonstrated that AI agents can now be utilized to reduce the computational cost of attacking 1024-bit RSA encryption keys, a development that could have long-term implications for data privacy and secure communications.
Beyond the realm of AI-specific threats, broader cybersecurity infrastructure remains under significant strain. The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has recently updated its Known Exploited Vulnerabilities catalog to include critical flaws in several widely used products. These include vulnerabilities in Check Point gateways, UnrealIRCd, IBM software, SignalWire FreeSwitch, OpenSSH, and GPUStack.
Specifically, unpatched Check Point gateways are currently being targeted by attackers who are wiping systems through the exploitation of PHP2Shell. Additionally, Microsoft’s November Patch Tuesday addressed a critical, exploited Outlook vulnerability alongside 30 other flaws categorized as wormable. In the mobile and desktop space, Zoom has also released a fix for a critical Windows vulnerability that allowed for unauthorized privilege escalation.
Data extortion remains a persistent threat, with several major organizations reporting large-scale compromises. A subsidiary of the French postal service, Maileva, confirmed a data breach affecting 22 million individuals. Meanwhile, an extortion group has claimed the theft of nearly one billion traveler records from Expedia, and the company WhiteArmor has issued a $5 million ransom demand to Maersk, alleging the theft of 100GB of proprietary data. These incidents continue to occur alongside state-level cyber activity, including the recent disruption of North Korean cryptocurrency theft operations, which saw the Department of Justice seize $1.2 billion in assets.





