Model Namespace Reuse Attack Demonstrated on Google, Microsoft AI Platforms

A critical vulnerability in Salesforce Agentforce, named ForcedLeak, allowed attackers to exfiltrate sensitive CRM data through indirect prompt injection. The flaw affected organizations using Salesforce Agentforce with Web-to-Lead functionality enabled. The vulnerability was discovered and reported by Noma Security on July 28, 2025. Salesforce has since patched the issue and implemented additional security measures, including regaining control of an expired domain and preventing AI agent output from being sent to untrusted domains. The exploit involved manipulating the Description field in Web-to-Lead forms to execute malicious instructions, leading to data leakage. Salesforce has enforced a Trusted URL allowlist to mitigate the risk of similar attacks in the future. The ForcedLeak vulnerability is a critical vulnerability chain with a CVSS score of 9.4, described as a cross-site scripting (XSS) play for the AI era. The exploit involves embedding a malicious prompt in a Web-to-Lead form, which the AI agent processes, leading to data leakage. The attack could potentially lead to the exfiltration of internal communications, business strategy insights, and detailed customer information. Salesforce is addressing the root cause of the vulnerability by implementing more robust layers of defense for their models and agents.

The UNC5221 activity cluster, attributed to suspected Chinese hackers, has been using the BRICKSTORM malware in long-term espionage operations against U.S. organizations in the technology, legal, SaaS, and BPO sectors. The malware, a Go-based backdoor, has been active for over a year, with an average dwell time of 393 days. It has been used to steal data from various sectors, including SaaS providers and BPOs. The attackers exploit vulnerabilities in edge devices and use anti-forensics techniques to avoid detection. The malware serves multiple functions, including web server, file manipulation, dropper, SOCKS relay, and shell command execution. It targets appliances without EDR support, such as VMware vCenter/ESXi, and uses legitimate traffic to mask its C2 communications. The attackers aim to exfiltrate emails and maintain stealth through various tactics, including removing the malware post-operation to hinder forensic investigations. The attackers use a malicious Java Servlet Filter (BRICKSTEAL) on vCenter to capture credentials, and clone Windows Server VMs to extract secrets. The stolen credentials are used for lateral movement and persistence, including enabling SSH on ESXi and modifying startup scripts. The malware exfiltrates emails via Microsoft Entra ID Enterprise Apps, utilizing its SOCKS proxy to tunnel into internal systems and code repositories. UNC5221 focuses on developers, administrators, and individuals tied to China's economic and security interests. Mandiant has released a free scanner script to help defenders detect BRICKSTORM. The BRICKSTORM backdoor is under active development, with a variant featuring a delay timer for C2 communication. The attackers have exploited Ivanti Connect Secure zero-day vulnerabilities (CVE-2023-46805 and CVE-2024-21887) for initial access. The attackers have used a custom dropper to install a malicious Java Servlet filter (BRICKSTEAL) in memory, avoiding detection. The attackers have modified init.d, rc.local, or systemd files to ensure persistence on appliances. The attackers have targeted Windows environments in Europe since at least November 2022. The attackers have been linked to other related Chinese threat actors besides UNC5221. The campaign has been monitored by Mandiant since March 2025. The attackers have targeted downstream customers of compromised SaaS providers. The attackers are believed to be analyzing stolen source code to identify zero-day vulnerabilities in enterprise technologies. The attackers use a delay timer to lie dormant on infected systems until a hard-coded date. The malware employs Garble, an open-source tool, for code obfuscation to hide function names, structures, and logic. Brickstorm has been found on VMware vCenter and ESXi hosts, often deployed prior to pivoting to these systems. The attackers use legitimate cloud services like Cloudflare Workers or Heroku for C2 communications. The attackers use dynamic domains like sslip.io or nip.io that point directly to the C2 server’s IP. The attackers favor appliance and management-plane compromise, per-victim obfuscated Go binaries, delayed-start implants, and Web/DoH C2 to preserve stealth. The attackers harvest and use valid high-privilege credentials to appear as routine administrator tasks. The attackers deploy in-memory servlet filters, remove installer artifacts, and embed delayed-start logic to limit forensic traces. The attackers abuse virtualization management capabilities, such as cloning VMs to extract credential stores offline. The attackers deploy an in-memory Java Servlet filter on vCenter to intercept and decode web authentication to harvest high-privilege credentials. The attackers use a SOCKS proxy on compromised appliances to tunnel into internal networks for interactive access and file retrieval.

A new attack vector, dubbed ShadowLeak, allows hackers to invisibly steal emails from users who integrate AI agents like ChatGPT with their email inboxes. The attack exploits the lack of visibility into AI processing on cloud infrastructure, making it undetectable to the user. The vulnerability was discovered by Radware and reported to OpenAI, which addressed it in August 2025. The attack involves embedding malicious code in emails, which the AI agent processes and acts upon without user awareness. The attack leverages an indirect prompt injection hidden in email HTML, using techniques like tiny fonts, white-on-white text, and layout tricks to remain undetected by the user. The attack can be extended to any connector that ChatGPT supports, including Box, Dropbox, GitHub, Google Drive, HubSpot, Microsoft Outlook, Notion, or SharePoint. The ShadowLeak attack targets users who connect AI agents to their email inboxes, such as those using ChatGPT with Gmail. The attack is non-detectable and leaves no trace on the user's network. The exploit involves embedding malicious code in emails, which the AI agent processes and acts upon, exfiltrating sensitive data to an attacker-controlled server. OpenAI acknowledged and fixed the issue in August 2025, but the exact details of the fix remain unclear. The exfiltration in ShadowLeak occurs directly within OpenAI's cloud environment, bypassing traditional security controls.

A vulnerability in the Cursor AI-powered Integrated Development Environment (IDE) allows automatic execution of tasks in malicious repositories upon opening. This flaw can be exploited to drop malware, hijack developer environments, or steal credentials and API tokens. The issue arises from Cursor disabling the Workspace Trust feature from Visual Studio Code (VS Code), which blocks automatic execution of tasks without explicit consent. This default behavior can be exploited by adding a malicious .vscode/tasks.json file in a publicly shared repository. The flaw affects Cursor's one million users who generate over a billion lines of code daily. The flaw can be exploited to leak sensitive credentials, modify files, or serve as a vector for broader system compromise, placing Cursor users at significant risk from supply-chain attacks. Cursor has decided not to fix the issue, citing the need to maintain AI and other features that depend on the autorun behavior. Users are advised to enable Workspace Trust manually or use a basic text editor for unknown projects.

A new variant of a TOR-based cryptojacking campaign targets exposed Docker APIs. The attack involves executing a new container based on the Alpine Docker image and mounting the host file system. The attackers then run a Base64-encoded payload to download a shell script downloader from a .onion domain. The script installs tools for reconnaissance and communication with a command-and-control (C2) server. The campaign may aim to establish a complex botnet. The attack chain includes exploiting additional ports (23, 9222) and using known default credentials for brute-forcing logins. The malware scans for open Docker API services at port 2375 and propagates the infection to those machines. The attackers block external access to port 2375 using available firewall utilities and install persistent SSH access. The malware includes dormant logic for future expansion opportunities for credential theft, browser session hijacking, remote file download, and distributed denial-of-service (DDoS) attacks. The campaign highlights the importance of securing Docker APIs and limiting exposure of services to the internet.