Deepfakes, Spyware Skits & LLMs for Hire

How Attackers Are Weaponizing Automation, Trust, and LLMs

This week’s headlines reinforce a few consistent patterns. Attackers are manipulating AI-driven workflows to trigger actions across SaaS platforms. Trusted marketplace integrations are becoming long-term attack paths. Advanced mobile spyware is becoming more accessible. And AI-assisted reconnaissance is shrinking the window between exposure and exploitation.

None of this represents a brand-new exploit class. What’s changing is where risk concentrates: in automation, third-party integrations, mobile endpoints, and response speed.

Key Takeaways:

Promptware Leveraging Google Calendar Invites
Cyber Security News

• What matters: Calendar invites aren’t new, but integrating AI assistants into systems that can take action introduces risk. Prompt injection exposure depends on the quality of the model and the surrounding software, and weaknesses in one implementation likely indicate weaknesses elsewhere.
• What’s overhyped: The delivery vector itself isn’t novel. This is an old technique being adapted to an AI-enabled environment, not a fundamentally new class of attack.

First Malicious Outlook Add-In Stealing Credentials
The Hacker News

• What matters: Domain takeovers can occur in places people don’t typically look, including multi-tenant Microsoft applications. In this case, the abandoned domain was tied to an Outlook add-in that applied across tenants and had permissions to read and write mail. The combination of multi-tenant exposure and privileged permissions increases impact. Third-party applications and add-ins require active review as part of risk management.
• What’s overhyped: Domain takeover itself isn’t new, and the core mechanism is familiar from subdomain takeover patterns. The difference here is where it occurred, not how it worked.

ZeroDayRAT Mobile Spyware Kit
SecurityWeek

• What matters: Mobile exploitation typically requires user interaction, such as installing an application, but this toolkit appears to leverage a browser-based vulnerability where simply visiting a webpage can establish access. The tooling is packaged as an all-in-one system with a dashboard for operators, enabling payload customization and deployment through phishing or coercion. It can escalate privileges, access sensors, GPS, camera, microphone, and other device data. More broadly, capabilities traditionally associated with nation-state actors are being packaged and commoditized for broader use, lowering the barrier to entry.
• What’s overhyped: Malware being sold via Telegram or underground channels is not new. The “nation-state grade” label depends on the level of compromise achieved; if it is user-level rather than kernel-level, it is materially different from tools like Pegasus. Many mobile compromises still depend on user behavior, such as installing untrusted applications, particularly in less restrictive app ecosystems.

AI-Assisted Deepfake CEO + Zoom Scam
PC Gamer

• What matters: Deepfake technology is being used not just for fraud, but to gain initial access into environments. The attack chain is consistent: fake calendar invite, Zoom call with real-time video or voice impersonation, and instructions delivered under the authority of executive leadership. In this case, the impersonation was used in a “ClickFix” scenario to persuade the victim to run commands, establish a foothold, and then pivot within the environment. The targeting focus remains cryptocurrency and finance-related organizations.
• What’s overhyped: The attack chain itself is not new. Similar executive impersonation campaigns using video occurred as early as 2024, following the same pattern of calendar invite, trusted call, and authority-based instruction. The difference is continued use and refinement, not a new methodology.

Poland Energy Grid Cyberattack & CISA Warning
Cyber Scoop

• What matters: The core issue is segmentation between corporate IT environments and operational technology environments. In this case, an exposed edge device connected to the internet lacked proper segmentation from the internal network, enabling lateral movement. As IT and OT systems converge, including cloud-managed device updates delivered over the air, historical separation weakens, increasing pivot risk.
• What’s overhyped: The pivot itself is not novel. The underlying issue remains basic network-level segmentation rather than advanced tradecraft.

Nation-State Hackers Using Gemini
The Record

• What matters: LLMs are being used by threat actors to automate reconnaissance, exploitation, and post-exploitation activities, including gaining situational awareness and pivoting laterally. Prior incidents have shown LLMs conducting these tasks across multiple organizations. Commercial frontier models provide some monitoring through API usage, but open-source and locally run models remove that oversight. The result is increased automation and operational efficiency for attackers.
• What’s overhyped: This is not the first instance of LLMs being used by adversaries. The development reflects continued automation and efficiency gains rather than a new category of attack.