View Ridge Security
Back to The Rollout

Microsoft disrupts EvilTokens platform linked to 12,000 account

Microsoft took down EvilTokens, an AI-powered cybercrime platform that compromised 12,000 accounts across 10,000 organizations.

🚀 RELEASES & LAUNCHES

  • Qualcomm unveils two smartphone chips built for on-device AI
    Qualcomm announced two smartphone chips that run AI models locally, including a 30B-parameter MoE model.

    • For smartphone makers and users seeking more on-device AI capabilities.
    • Snapdragon 8 Elite Gen 6 and 8 Elite Extreme Gen 6 support sensing-hub models up to 200M parameters and local personal-scribe features.
    • The Extreme chip runs a 30B-parameter mixture-of-experts model locally and supports 8K 60fps and 4K 240fps video.
    • The chips add local voice agents, speaker differentiation, usage-based memory, AI noise reduction, and voice isolation for calls.
      📎 https://techcrunch.com/2026/09/22/qualcomm-launches-two-new-smartphone-chips-with-emphasis-on-ai/ · 👁 via TechCrunch AI

  • Rabbit rolls out OS3 AI agent for Windows, Mac, and Linux
    Rabbit is rolling out its standalone OS3 AI agent without requiring an R1 device.

    • OS3 is for Windows, Mac, and Linux users, with access also available through a desktop site, messaging apps, and the R1.
    • The cloud-based agent operates locally, supports up to five devices per account, and can use preferred AI models.
    • OS3 selects the devices, files, apps, and models needed for each task, while Rabbit says it will not store, copy, use, or sell user data.
    • Rabbit says chats and memories remain on its servers, and linked AI providers apply their own privacy policies.
      📎 https://www.theverge.com/ai-artificial-intelligence/999094/rabbit-ai-agent-os3 · 👁 via The Verge AI

📊 BENCHMARKS & EVALS

  • Anthropic and OpenAI launch lower-cost frontier models
    Anthropic and OpenAI released lower-cost frontier models for enterprise AI use.
    • Developers and enterprise AI teams are the target users for the new models.
    • Anthropic’s Claude Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, with cache reads at $0.20 per million and output generation over 30% faster than Opus 5.
    • OpenAI’s GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens; GPT-6 Luna costs $0.10 and $0.50, respectively.
    • The releases give builders cheaper options as enterprises compare frontier models with open-weight systems and model routers.
      📄 https://www.anthropic.com/claude-opus-5-5 · 📎 https://arstechnica.com/ai/2026/09/new-anthropic-openai-models-make-same-promise-a-little-more-for-a-lot-less-money/ · 👁 via Ars Technica AI

💰 BUSINESS & FUNDING

  • Snorkel AI raises $350M at a $3.5B valuation
    Snorkel AI raised a $350 million Series E at a $3.5 billion valuation.
    • AI labs and corporations use Snorkel AI to build training datasets and simulated environments.
    • The startup now sells complete datasets and reinforcement-learning environments through a data-as-a-service model.
    • Insight Partners and S32 led the Series E, which nearly tripled Snorkel’s valuation from $1.3 billion.
    • Snorkel reported a $375 million annualized revenue run rate, up 18-fold in 12 months, reflecting demand for specialized AI training data.
      📎 https://techcrunch.com/2026/09/22/snorkel-ai-triples-valuation-to-3-5b-as-demand-for-ai-training-data-booms/ · 👁 via TechCrunch AI

⚠️ INCIDENTS & SAFETY

  • Microsoft disrupts EvilTokens platform linked to 12,000 account compromises
    Microsoft disrupted EvilTokens, an AI-assisted cybercrime platform linked to 12,000 account compromises.
    • EvilTokens targeted cybercriminal subscribers with a $1,500 setup fee and $500 monthly charge.
    • The platform used an AI chatbot to analyze up to 5,000 compromised emails, identify high-value targets, and draft convincing fraud messages.
    • Microsoft seized 50 websites and 150 additional domains; UK police arrested two men over alleged links to the platform.
    • EvilTokens compromised 12,000 accounts across 10,000 organizations, including firms in finance, healthcare, construction, education, and real estate.
    • The AI-assisted workflow reduced inbox analysis and fraud preparation from days to minutes for attackers.
      📄 https://blogs.microsoft.com/on-the-issues/2026/09/22/disrupting-eviltokens-the-ai-chatbot-built-for-cybercrime/ · 📎 https://arstechnica.com/security/2026/09/microsoft-disrupts-ai-assisted-platform-that-compromised-12000/ · 👁 via Ars Technica AI

🏛️ POLICY & REGULATION

  • UPDATE: British Columbia sues OpenAI over Tumbler Ridge shooting response
    British Columbia sued OpenAI over the Tumbler Ridge shooting.
    • British Columbia and a local school board are suing OpenAI and Sam Altman.
    • The province seeks ChatGPT logs, replacement-school costs, emergency expenses, and safety changes.
    • The shooting killed eight people, forced the school’s demolition, and left about 160 students and staff traumatized.
    • The lawsuit asks ChatGPT to automatically terminate violent conversations and seeks public disclosure of the logs.
      📄 https://cdn.arstechnica.net/wp-content/uploads/2026/09/British-of-Columbia-v-Altman-Complaint-9-21-26.pdf · 📎 https://arstechnica.com/tech-policy/2026/09/lawsuit-demands-openai-pay-for-new-school-after-chatgpt-used-in-shooting/ · 👁 via Ars Technica AI

Make AI adoption safer

Build practical guardrails around the AI tools your team already uses.

Explore Secure AI Adoption