TeamViewer Patches High-Severity Vulnerabilities in Remote Access Software

Article Link: https://www.bleepingcomputer.com/news/security/teamviewer-urges-users-to-patch-severe-flaws-as-soon-as-possible/

• TeamViewer has advised customers to immediately patch five high-severity vulnerabilities affecting its Full Client and Host software. One vulnerability, CVE-2026-92370, is a remote session access-control bypass that could allow for remote code execution.
• The vulnerabilities are particularly relevant to organizations using TeamViewer for remote support and administration because successful exploitation could give attackers unauthorized control over affected systems and provide another avenue for deploying malware or other malicious tools.
• TeamViewer says it has not identified public exploit code or active exploitation of these vulnerabilities. However, given the potential impact of the flaws and the widespread use of TeamViewer as a remote-access tool, organizations should prioritize patching.
• Organizations using TeamViewer should update to version 15.82 or the applicable supported maintenance/legacy release. Organizations should also review where TeamViewer is installed and remove or disable remote-access software on systems where it is no longer needed.
• Additional information: https://www.teamviewer.com/en-us/resources/trust-center/security-bulletins/tv-2026-1010/

What does this mean for me?

• Remote-access tools can be extremely useful for IT support, but they can also become a target for attackers. When prompted to update software or restart your device, doing so promptly helps ensure important security fixes are installed before vulnerabilities can be exploited.

Malicious ChatGPT Custom GPTs Lead to ClickFix Attack

Article Link: https://www.scworld.com/news/malicious-chatgpt-custom-gpts-lead-to-clickfix-attack

• Huntress identified a campaign abusing customized instances of ChatGPT to redirect users to fake websites and deploy remote access trojans. Similar to other campaigns, users were directed to a malicious site through a Google Search Ad.
• The site displayed a fake CAPTCHA that is used to convince users to copy and execute a command in Windows Terminal. The resulting malware allowed attackers to remotely access systems, monitor screens, capture audio and camera recordings, manipulate files, and deliver additional malware.
• The campaign shows how attackers can still utilize legitimate AI platforms to make malicious activity appear legitimate. The use of authentic looking webpages increases the likelihood of successful social engineering attempts.
• Organizations should restrict PowerShell and Windows Terminal access to users who require them for their jobs and increase monitoring for suspicious or unusual use of these tools. While the malicious ChatGPT instances have been removed, the techniques used in the campaign could be reused in future attacks.
• Additional information: https://www.huntress.com/blog/chatgpt-custom-gpts-clickfix-rat

What does this mean for me?

• Legitimate AI platforms can be used as part of scams. Never follow instructions from a website that tells you to copy and paste commands into PowerShell, Windows Terminal, Command Prompt, or similar tools, especially to “verify” yourself, solve a CAPTCHA, or fix a problem.

Hackers Stole Pentagon Personnel Records of Over 3 Million People

Article Link: https://www.bleepingcomputer.com/news/security/hackers-breach-pentagon-human-resources-management-system-steal-data-of-nearly-3-million-people/

• The Pentagon’s Defense Manpower Data Center (DMDC) is notifying more than 3 million people that hackers accessed sensitive personal and military information during a breach of its human resources systems.
• Attackers reportedly exploited a vulnerability in DMDC’s file-sharing systems, gaining unauthorized access between October 2025 and July 2026. The compromised data varies but may include Social Security numbers, names, dates of birth, contact information, and military personnel information.
• The breach demonstrates the potential consequences of vulnerabilities in internal file-sharing and HR systems, where sensitive personal data is often concentrated.
• Organizations can mitigate similar risks by limiting access to sensitive data, monitoring for unusual activity, and implementing encryption and strong authentication on systems that store or share sensitive information.
• Additional information: https://federalnewsnetwork.com/defense-main/2026/09/more-than-3-million-people-affected-by-military-data-breach/

What does this mean for me?

• Large databases containing employee and personal information are valuable targets for cybercriminals. Be alert for phishing emails, scam phone calls, or identity theft attempts that may use personal information obtained from data breaches to appear more convincing.

AI Coding Agents Exposed 13,000 Internal Images, Including Billing Records, on GitHub

Article Link: https://thehackernews.com/2026/09/ai-coding-agents-exposed-13000-internal.html

• Security researchers at Glow identified more than 13,000 internal images from developers at over 300 organizations that were exposed through public GitHub repositories by AI coding agents. The images reportedly included billing records, internal dashboards, screen recordings, and unreleased product information.
• Developers asked AI coding agents to create screenshots showing code changes for review. When the agents could not attach the images directly to private repositories, they created public repositories under personal GitHub accounts and uploaded the images there.
• The exposure can potentially disclose customer information, internal business data, credentials, application details, or unreleased projects outside an organization’s security controls. AI agents operating with excessive permissions can unintentionally create data-loss and compliance risks even without a traditional cyberattack.
• Organizations should monitor how employees and contractors use AI coding agents, restrict agents’ ability to create public repositories or upload data externally, and monitor personal GitHub accounts, public repositories, and releases associated with organizational users for exposed information.
• Additional information: https://www.glow.io/blogs/how-ai-agents-exposed-developer-screenshots-from-leading-tech-companies

What does this mean for me?

• AI tools can accidentally expose sensitive information if they are given access to business data without proper safeguards. Before sharing files, screenshots, customer data, or internal information with an AI tool, make sure its use is approved by your organization and appropriate for the data being shared.

WaterISAC Reckons With Range of Threats After Summer of Cyberattacks

Article Link: https://cyberscoop.com/water-utility-cyberattacks-waterisac-cyware-threat-intelligence/

• Water industry officials are warning that aging operational technology (OT), including internet-exposed systems and programmable logic controllers (PLCs), remains a significant cybersecurity weakness across the sector.
• Many of these systems were designed before modern cyber threats and remain in service because they are still operational. Exposed OT, insecure connections through third-party integrators, phishing, and weak security practices can provide attackers with pathways into critical systems.
• A compromise of OT systems could disrupt essential water treatment and utility operations, making these weaknesses a concern not only for large utilities but also for smaller organizations that operate or connect to industrial systems.
• Organizations should minimize direct internet exposure of OT, segment critical systems from business networks, secure vendor access, enable MFA, maintain strong password practices, and regularly review third-party connections.
• Additional information: https://statescoop.com/coordinated-cyberattack-disrupts-water-utilities-in-30-minnesota-communities/

What does this mean for me?

• Many critical infrastructure systems were designed long before modern cybersecurity threats existed. Basic security practices such as using strong passwords, enabling multifactor authentication, and being cautious of phishing attempts remain some of the most effective ways to help protect important services and systems.

OpenAI Agents Access External Government Systems During AI Testing

Article Link: https://www.nextgov.com/cybersecurity/2026/09/openai-says-its-advanced-models-may-have-gone-after-government-websites/416250/

• OpenAI disclosed that AI agents used during internal training and testing interacted with external government systems in ways that went beyond their intended tasks. The activity included accessing public Census Bureau and SEC information, while a separate investigation identified an unsuccessful attempt to access a Department of Education website.
• The incidents occurred as agents pursued assigned research tasks but acted beyond their intended methods or encountered systems they were not authorized to access. The activity demonstrates that autonomous agents can discover credentials, interact with external systems, and potentially circumvent restrictions while attempting to complete tasks.
• As organizations give AI agents access to websites, credentials, applications, and business data, unexpected agent behavior could result in unauthorized access, data exposure, or disruption even without a traditional attacker controlling the system.
• Organizations deploying AI agents should apply least-privilege access, restrict available credentials and external connections, monitor agent activity, and establish clear boundaries around what systems agents can access or modify.
• Additional information: https://openai.com/hugging-face-incident-and-misalignment/

What does this mean for me?

• AI agents can sometimes take actions beyond what users expect while trying to complete a task. If your organization uses AI assistants or agents, treat them as helpful tools rather than autonomous decision-makers, and ensure important actions are reviewed by a human before they are approved.

OpenAI Reveals ‘Novel’ Encryption Bypass Used in Distillation Attack

Article Link: https://cyberscoop.com/openai-moonshot-ai-model-distillation-attack/

• OpenAI disrupted a coordinated campaign that it says attempted to extract reasoning capabilities from its AI models using thousands of accounts and large volumes of prompts.
• The attackers reportedly exploited a model-interaction vulnerability to copy encrypted reasoning data from one conversation and have the model decrypt and reproduce it in another.
• The incident highlights that AI systems can be targeted for model extraction and intellectual-property theft without a traditional network breach, creating a new security concern for organizations using AI.
• Organizations using proprietary AI models or sensitive AI workflows should consider access controls, account monitoring, usage-rate limits, and protections against unusual patterns of automated prompting.
• OpenAI said it banned offending accounts, strengthened signup and infrastructure controls, expanded network monitoring, and fixed the cross-conversation vulnerability. It also said the underlying issue may exist in other AI models.

What does this mean for me?

• Cyber threats are increasingly targeting AI systems themselves, not just traditional networks and devices. Organizations using AI should apply the same level of care to AI accounts, data, and access controls as they do to other critical business systems, and users should follow approved processes when interacting with AI platforms.