Disrupting EvilTokens: Taking down an AI-powered cybercrime platform

blogs.microsoft.com Sep 23, 2026

Disrupting EvilTokens: Taking down an AI-powered cybercrime platform

Curated by Shubham · TechCapsules · Sep 23, 2026

AI-summarised brief · reviewed before publication

Microsoft announced the takedown of EvilTokens, an AI‑driven cybercrime platform that automated the full fraud pipeline from email compromise to financial theft. Launched in February 2026, the service sold access to compromised inboxes for a $1,500 initiation fee and $500 monthly, using a chatbot to scan victims’ mail, map organizational roles, pinpoint “money movers,” and draft convincing impersonation messages. Within months it breached more than 12,000 inboxes across over 10,000 organizations in the United States, Canada, the United Kingdom, Australia, India and France, affecting sectors from finance to healthcare. Microsoft, in partnership with law‑enforcement and hosting providers, seized 50 websites, disabled 150 domains, and supported the arrest of two suspects in the UK. The operation highlighted how AI can lower the expertise barrier for sophisticated fraud.

💡 Why It Matters

· EvilTokens proves that AI can turn routine credential theft into a turnkey financial‑fraud service, dramatically expanding the pool of capable attackers.

AI Capsules Cybersecurity Microsoft

Read full article on blogs.microsoft.com

Nest Doorbell has been triggering Gemini’s American accent, no matter what language…

Google Home users report that Nest Doorbell notifications trigger Gemini’s voice in an American accent, regardless of the user’s language settings. The glitch, first noted by a French resident, also affects Dutch and UK users, with appliances like Bosch washing machines defaulting to US English. Gemini otherwise speaks in the chosen language, but the Nest Doorbell and similar devices consistently use an American accent. Google acknowledges the problem and is working on a fix, with no release date yet.

💡 The bug exposes a flaw in Google’s multilingual integration, risking user confusion and undermining trust in voice‑assistant consistency across global markets.

therobotreport.com · Sep 23

Qualcomm to acquire PickNik Robotics and keep MoveIt open-source

Qualcomm Technologies announced the acquisition of PickNik Inc., the creators of the MoveIt open‑source robotics framework. The deal will merge Qualcomm’s Dragonwing AI‑powered hardware with MoveIt’s manipulation software, aiming to simplify the pipeline from AI models to real‑time robotic control. PickNik, founded in 2015, has partnered with major clients such as Google and NASA, and recently released MoveIt Pro 9.0. Qualcomm commits to keeping MoveIt 1 and 2 open source, supporting community collaboration and cross‑platform compatibility.

💡 The integration accelerates the deployment of AI‑driven robots across industries, bridging advanced software with Qualcomm’s scalable hardware. It reinforces an open‑source ecosystem that lowers entry barriers for developers and speeds innovation in physical AI.

thehackernews.com · Sep 23

This Windows Malware is Built to Let Up to Four AI Models…

Cisco Talos uncovered a Windows malware named CLOSEDQUORUM that delegates command decisions to up to four commercial AI services—DeepSeek, Qwen, Mistral, and Google Gemini—instead of a traditional command‑and‑control server. The code, dated June 17, 2026, instructs the AI models to choose from four actions: steal credentials, inject code, persist, or move. The malware reports its choices and stolen data to an attacker‑controlled Discord channel via a webhook. The public sample contains placeholder API keys, rendering it non‑functional, but Talos confirmed the design and potential impact.

💡 By outsourcing attack logic to AI, CLOSEDQUORUM introduces a novel, decentralized threat model that can evade conventional detection tied to known C2 domains. This approach forces defenders to monitor anomalous AI‑service interactions and internal Windows behaviors rather than relying solely on domain blacklists.

9to5google.com · Sep 23

YouTube Music adding corner access to ‘Ask Music’ with Gemini

At Made on YouTube 2026, Google unveiled several YouTube Music enhancements slated for release later this year. The most visible change moves the Gemini‑powered “Ask Music” shortcut from a Home‑feed card to a permanent spark icon in the app’s top‑right corner, opening a sliding chat sheet with text and voice input. Suggested prompts let users create mixes, discover artists or find podcasts, and the tool will appear contextually when editing playlists or building mixes, offering deeper artist information and queue customization. A second addition, “Your Podcast Lineup,” delivers a weekly, AI‑generated audio briefing that explains recommended shows and episodes, providing a hands‑free discovery experience. Both features will roll out globally to YouTube Premium and Music Premium subscribers through the platform.

💡 By embedding Gemini‑driven chat directly into core navigation, YouTube Music turns passive browsing into an interactive, voice‑first experience, nudging users toward deeper engagement with its catalog.

citynewsservice.cn · Sep 23

Daily Buzz: 23 September 2026

Alibaba Group announced an ambitious AI roadmap at its Apsara Cloud conference in Hangzhou, targeting the development of models with 5‑10 trillion parameters. Chief executive Eddie Wu detailed the progression from the current Qwen 3.8 Max (2.4 trillion parameters) to the in‑training Qwen 4 and future Qwen 4.5 and Qwen 5 series, which aim to handle more complex, longer‑horizon tasks and move toward superintelligence. The company highlighted its proprietary M890 AI supernode, capable of inference for models exceeding 2 trillion parameters, a capability it claims only a few firms possess. Alibaba also unveiled the Zhenwu V900, a next‑generation AI chip from its T‑Head unit, slated for mass production and commercial release in Q1 2027. Wu emphasized that machine cognition, now under 3 percent of total thinking, could expand dramatically.

💡 Alibaba’s push to scale AI models and chips positions it as a rare global contender in ultra‑large model deployment, potentially reshaping competitive dynamics in China’s AI hardware and cloud services markets.

indianexpress.com · Sep 23

Ex-Google safety chief warns AI could harm children more than social media…

Former Google trust‑and‑safety vice president Tom Siegel warned Reuters that artificial‑intelligence tools could damage children more severely than social media ever did. Siegel, who led Google’s safety team, said the industry is repeating past mistakes and cannot rely on self‑regulation. He highlighted recent revelations, such as a Meta internal memo describing chatbots that could engage minors in romantic or sensual dialogue, prompting a congressional probe. While governments are already restricting minors’ access to social platforms, AI development is outpacing existing rules. Siegel announced his new role as executive director of the Youth AI Safety Institute, a research arm of the nonprofit Common Sense Media, which evaluates media suitability for kids. He called for external oversight to prevent further harm.

💡 Without independent regulation, AI’s rapid integration into youth‑focused apps could expose a generation to unprecedented psychological risks.