Major Data Breaches
Dropbox Account Compromises via Lenovo ID
During the week, Dropbox disclosed that approximately 5,000 user accounts were compromised after attackers exploited an authentication flaw involving Lenovo ID single sign-on integrations. The attackers abused an email verification weakness on Lenovo’s identity platform, registering external accounts with victims’ corporate email addresses. Dropbox’s federated login configuration mistakenly accepted the asserted email identity without requiring a secondary password challenge, allowing unauthorized access and download of user files from connected accounts lacking independent two-factor authentication. Dropbox responded by invalidating all active sessions linked through Lenovo ID and terminating the legacy integration. This incident highlights the risks of automatic account linking based solely on shared email addresses without explicit cryptographic or administrative confirmation1.
Significant Cyberattacks
Autonomous AI Agents Breach Enterprise Network
Palo Alto Networks’ Unit 42 reported a sophisticated attack in which adversaries used autonomous AI agents to compromise an enterprise network in under ten hours. The operation automated more than fifty MITRE ATT&CK techniques, including internal reconnaissance, secret discovery across code repositories, and master credential harvesting from centralized secrets managers. The AI agents compromised CI/CD pipelines to extract cloud keys, attempted to inject backdoors into Terraform templates, and generated a comprehensive technical audit of the target environment for extortion leverage. Investigators identified structured Markdown handoffs between parallel agent sessions as clear indicators of agentic orchestration. Defenders are advised to implement strict code review policies and automated credential revocation mechanisms to counter such fast-paced intrusions1.
RevStealer Targets Claude Sessions
Security analysts revealed that cybercriminals are increasingly targeting authenticated web session cookies and authentication tokens belonging to Anthropic Claude accounts. Specialized information-stealing malware extracts stored browser cookies from infected employee machines, allowing attackers to bypass multi-factor authentication and establish persistence within organizational AI workstreams. Once inside an active Claude session, attackers can access sensitive internal conversations, proprietary code snippets, and confidential prompt histories. Organizations are urged to enforce short session timeouts, deploy endpoint protections against infostealers, and educate staff about the risks of pasting sensitive proprietary secrets into conversational interfaces1.
Critical Vulnerabilities
Google Chrome V8 Zero-Day Actively Exploited
Google released an emergency update to address a high-severity zero-day vulnerability in its V8 JavaScript engine, tracked as CVE-2026-85046. This type confusion flaw, discovered by security researcher Salvatore Gulizia, allows memory corruption or arbitrary code execution within the browser process. The patch advances Chrome Stable to version 152.0.7977.82/.83 on Windows and macOS, and 152.0.7977.82 on Linux. Exploitation typically requires luring victims to malicious web pages via phishing, malvertising, or compromised sites. Administrators are urged to expedite browser updates across managed endpoints2.
Magento and Adobe Commerce StyleSmuggler Zero-Day
Sansec uncovered an actively exploited, unauthenticated remote code execution zero-day vulnerability dubbed StyleSmuggler affecting Magento Open Source and Adobe Commerce, including version 2.4.9. Attackers manipulate style properties inside unauthenticated GraphQL queries to smuggle PHP code into log and report files, triggering code execution when Magento renders its standard payment failure notification email. The exploit deploys a persistent Rust binary disguised as a legitimate kernel thread. With no official patch released, administrators are advised to temporarily disable GraphQL and restrict process execution permissions2.
MikroTik RouterOS Flaws Exploited
CERT Polska warned of active exploitation of two zero-day flaws in MikroTik RouterOS, codenamed MikroTrick. The vulnerabilities (CVE-2026-67276 and CVE-2026-86060, CVSS scores: 9.2) allow attackers to bypass authentication and elevate privileges if the device supports remote access via SSH. Successful attacks have been observed since at least September 2, originating from specific IP addresses. MikroTik has released fixes in versions 6.49.21 (Long-term), 7.23.4 (Long-term), and 7.24.2 (Stable). Administrators should update immediately23.
N-able N-central Critical Flaws
N-able released hotfixes for two severe N-central flaws (CVE-2026-86206 and CVE-2026-86207) that could allow unauthorized parties to bypass authentication controls and gain full access to the platform. A maximum-severity flaw (CVE-2026-86218, CVSS score: 10.0) could allow pre-authenticated remote code execution on the N-central server. While no confirmed exploitation in production environments has been reported, Huntress observed signs of likely exploitation following a compromise of a fully patched N-central production environment. Administrators are strongly advised to patch immediately2.
VMware Workstation and Fusion Host Code Execution
Broadcom issued advisory VMSA-2026-0007 detailing two security flaws in VMware Workstation and Fusion. The most severe, CVE-2026-59346 (CVSS score: 9.3), is an integer overflow in the VMXNET3 virtual network adapter, allowing attackers with local administrative privileges inside a guest VM to execute arbitrary code on the host OS. CVE-2026-59347 (CVSS score: 8.1) is a stack-based buffer overflow in the Host-Guest File System shared folders component. Both issues are addressed in version 26H1u1, and immediate patching is recommended1.
Government Responses
CISA Adds Exploited Flaws to KEV Catalog
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) added five security flaws impacting JFrog Artifactory, ConnectWise ScreenConnect, and MikroTik RouterOS to its Known Exploited Vulnerabilities (KEV) catalog following reports of active exploitation. Details include CVE-2026-42016 (JFrog Artifactory, CVSS score: 8.1), CVE-2026-42018 (JFrog Artifactory, CVSS score: 7.5), and CVE-2026-84869 (ConnectWise ScreenConnect, CVSS score: 9.9). CISA’s action underscores the urgency for organizations to prioritize patching these vulnerabilities4.
OpenAI Pledges $1B for Cyber Defense
OpenAI announced a $1 billion initiative to provide resources and training for frontline cyber defenders, leveraging frontier AI to help water, power, and local government providers combat malicious actors. The company aims to use advanced AI models to accelerate patch development and defensive measures, while also addressing dual-use concerns regarding the democratization of sophisticated exploit engineering5.
Miscellaneous
OpenAI GPT-6 Astra Discovers Zero-Days
OpenAI introduced GPT-6 Astra, a frontier intelligence model capable of identifying software zero-day vulnerabilities and constructing functional proof-of-concept exploits during authorized offensive security benchmarks. Astra achieved a 100% score on ExploitBench, demonstrating advanced autonomous computer-use and debugging capabilities. While automated flaw discovery accelerates patch development, the release highlights dual-use concerns as adversaries may leverage similar capabilities1.
Anthropic Disrupts Industrial-Scale Claude Distillation Attacks
Anthropic identified and disrupted industrial-scale illicit distillation attacks against Claude from seven labs based in China, including Alibaba, Moonshot, DeepSeek, Z.ai, and MiniMax. Illicit distillation covertly extracts a model’s capabilities and replicates them in another model without authorization, typically using networks of fake accounts created with stolen credit cards, login credentials, and API keys. Anthropic’s actions highlight the growing threat of AI model theft and the need for robust protections4.
Conclusion
The week of September 7–13, 2026, saw a surge in critical vulnerabilities, sophisticated cyberattacks leveraging AI, and significant data breaches affecting major platforms. Government agencies and industry leaders responded with urgent advisories and new initiatives, emphasizing the need for rapid patching, improved identity controls, and advanced defensive strategies. As AI continues to transform both attack and defense landscapes, organizations must adapt quickly to evolving threats and prioritize resilience alongside prevention.
Sources: