The cybersecurity landscape continues to shift rapidly as organizations grapple with the dual challenges of securing generative AI infrastructure and defending against increasingly sophisticated, automated attack vectors. This week’s roundup covers critical developments across the threat intelligence spectrum, ranging from vulnerabilities in AI deployment frameworks to significant data breaches and a marked rise in the velocity of exploits targeting enterprise software.
Anthropic Enhances Claude Enterprise Security
Anthropic has taken significant steps to address the security needs of its corporate clientele by introducing a suite of robust administration and security capabilities for Claude Enterprise. The company’s focus is on providing organizations with the granular control necessary to safely integrate large language models (LLMs) into their workflows.
A centerpiece of this update is a new security center, currently in beta, which provides automated configuration audits. This tool evaluates an organization’s setup—including chat configurations, Cowork, and Claude Code—against Anthropic’s internal security best practices. Furthermore, the general availability of inference hooks now allows enterprises to route prompts and tool calls directly into existing data loss prevention (DLP) and security monitoring tools. This ensures that sensitive data is inspected in real-time before being processed by the model. Complementing these tools is a centralized management console, also in beta, which streamlines the administration of single sign-on (SSO) and SCIM protocols across multiple Claude organizations, providing a unified security posture.
Sophisticated Phishing Campaigns Target AI Event Registrants
The allure of emerging AI technology is being weaponized by threat actors to compromise corporate credentials. Push Security recently documented a deceptive phishing campaign that mimics event registration pages for major industry players, including OpenAI, SpaceX, and Anthropic. The campaign baits marketing professionals with promises of exclusive rewards, such as extended trials for ChatGPT Pro or Grok Heavy.
When a user attempts to sign up, they are prompted to authenticate using their corporate Microsoft or Google credentials. The Microsoft-targeted flow is particularly concerning, as it utilizes the Kali365 phishing-as-a-service platform to execute a device code phishing attack. Researchers have linked this activity, with medium-high confidence, to previous campaigns involving fraudulent Zoom event pages. To maintain operational longevity, the attackers are employing a rotating infrastructure of short links from Google and Zapier, paired with Cloudflare challenges and frequently swapped domains, all designed to evade detection by traditional security filters.
Chainalysis Breach Disclosure Widens
The transparency surrounding the security incident at blockchain analysis firm Chainalysis has intensified following a revised disclosure. While the company initially reported that only non-sensitive data had been compromised, a follow-up announcement on November 2 revealed a much broader impact. Chainalysis confirmed that the breach, which was first reported after an AI-driven cyberattack, exposed names, usernames, password hashes, and live tokens for its entire user base.
In response to the severity of the data exposure, the company has taken several products offline, most notably its KYT (Know Your Transaction) monitoring service, after detecting evidence of abnormal activity. Chainalysis is currently in the process of notifying customers whose accounts show clear indicators of unauthorized access, marking a significant escalation in the scope of the incident.
Google Bolsters AI Security with CNCF Donations
In a move aimed at establishing a vendor-neutral standard for AI security, Google has donated two of its critical projects, Agent Substrate and gVisor, to the Cloud Native Computing Foundation (CNCF) as Sandbox projects. As organizations scale their use of autonomous AI agents, the risk of untrusted code execution has become a primary concern.
Agent Substrate provides an orchestration layer for Kubernetes, designed to manage isolated AI agent sessions at scale. Its ability to save the state of an idle agent to storage and reactivate it in under 500 milliseconds makes it highly efficient for production environments. Complementing this, gVisor acts as an application kernel that intercepts system calls in user space, creating a secure boundary that prevents AI-generated code from interacting directly with the host kernel. By moving these projects under the CNCF, Google aims to provide the developer community with a standardized, secure infrastructure for running agents across any Kubernetes-compliant environment.
The ‘Greeter’ Framework Targets Flowise Vulnerabilities
Security researchers at Era have uncovered a new, Chinese-language exploitation framework dubbed "Greeter," which has been actively probing Flowise AI infrastructure. The framework employs a high-volume, automated approach, bombarding exposed instances with thousands of messages containing only the word "hello."
Behind this repetitive, seemingly benign traffic lies a sophisticated effort to exploit a known code injection vulnerability (CVE-2024-58351). Once the framework establishes a foothold, it attempts to map the host environment, query cloud metadata services, and harvest SSH keys or other credentials. It also tests for the capability to execute payloads directly from memory, a technique used to evade disk-based detection. Era’s analysis revealed that of the over 4,300 Flowise instances exposed online, hundreds were either unauthenticated or configured in a manner identical to the vulnerable honeypots used in the researchers’ study.
LLM Jacking and the Rise of ‘LLM Mixers’
The monetization of hijacked enterprise AI access has evolved into a specialized underground economy, according to investigations by Mandiant. This phenomenon, known as "LLMJacking," involves attackers seizing control of cloud-based AI model access to perform unauthorized, high-cost queries. In some instances, these intrusions have resulted in fraudulent billing reaching over $100,000 in a single day, with total losses at some organizations exceeding $1 million.
Initial access is frequently gained through stolen or exposed cloud credentials and API keys. In one notable case, attackers leveraged a vulnerability in LiteLLM (CVE-2026-40217) to escalate standard API keys into full administrative credentials. Beyond direct exploitation, Google’s Threat Intelligence Group has identified the rise of "LLM Mixers," such as the Jet Night token routing service. These entities aggregate stolen keys to offer "unlimited" access to frontier models for as little as $60 per month, effectively laundering stolen corporate compute resources for profit.
Security Concerns Surround CryptoBilis Hardware Wallets
Trezor has issued an urgent warning to customers who purchased hardware wallets from the Southeast Asian reseller CryptoBilis since March 2026. The alert follows a report from a customer who discovered a malicious implant inside a Trezor Safe 5 device. The hardware modification allegedly consists of an added circuit board equipped with a cellular module and antenna, designed to capture the recovery seed phrase as it is displayed on the device screen and transmit the data over a mobile network.
While Trezor has not yet independently verified the report, the potential for such an attack is significant, as the same reseller was previously implicated in the sale of tampered Ledger devices. As a precautionary measure, Trezor is advising all affected customers to move their assets to a freshly generated wallet with a new, secure seed phrase. To date, no specific reports of stolen funds related to these tampered units have been confirmed.
Impact of the 2025 Harvard Data Breach
The scale of the 2025 security incident at Harvard University has been further clarified by its inclusion in the "Have I Been Pwned" database. Approximately 851,000 unique email addresses associated with the breach have been indexed, confirming the depth of the data theft. The records, which originated from the university’s alumni and fundraising systems, include sensitive personal profiles.
Information leaked in the breach, which has been attributed to an extortion bid by the group known as ShinyHunters, includes names, contact details, birth dates, employment history, ethnicity, marital status, and detailed donation records. The exposure of this information poses a significant risk of targeted phishing and social engineering attacks against the affected alumni, donors, and staff members.
Escalating Exploitation Speeds in SAP Environments
The window for patching vulnerabilities in enterprise software is closing rapidly, particularly for SAP environments. According to a joint report by Onapsis and Recorded Future, the median time between the release of a security patch and the first verified exploitation or public proof-of-concept for SAP vulnerabilities has shrunk to just one to three days. This is a dramatic decrease from the 489-day average observed for flaws patched in 2021.
This trend is driven by an increase in the number of ransomware and extortion groups targeting SAP platforms, which grew from 8 in 2023 to 26 in 2025. Furthermore, the number of SAP CVEs with confirmed in-the-wild exploitation has doubled in less than three years. The research underscores the professionalization of the underground market, where access to vulnerable SAP systems is now being listed for prices ranging from $500 to $30,000.
Synology Urges Action on Outdated DSM Versions
Synology has issued a warning regarding active exploitation attempts targeting a vulnerability fixed in late 2024. The flaw, referenced in advisory Synology-SA-24:20, is currently being leveraged by attackers to target devices running outdated versions of the DiskStation Manager (DSM) operating system. The objective of these attacks is to gain unauthorized access to the network-attached storage (NAS) units, encrypt stored data, and extort victims for ransom. Synology has strongly urged all users to verify their firmware versions and enable automatic updates to ensure that the security patch is applied, thereby mitigating the risk of ransomware infection.
Nomad Bridge Exploiter Extradited
Finally, the legal repercussions for the 2022 exploit of the Nomad Bridge have moved forward with the extradition of Alexander Gurevich (also known as Alexander Block) from Israel to the United States. Gurevich made his first appearance in a U.S. court on October 30 to face charges related to the theft of approximately $2.89 million in virtual currency. Prosecutors allege that after exploiting the bridge’s smart contract, the suspect laundered the stolen funds by swapping WBTC and WETH across multiple obfuscated addresses. The charges include money laundering, a crime that carries a potential sentence of up to 20 years in federal prison.

