Apple Opens Siri AI; Google Ties Search to AI; CrowdStrike's Kill Switch
Apple has opened Siri AI to third-party developers, while Google's new school policy ties Search access to AI settings. CrowdStrike introduced an AI kill switch for enterprise agents, and MIT researchers demonstrated AI models improving others by rewriting weights.
Podcast В· 2 min
Apple opens Siri AI to third-party developers
Apple has released a developer preview of its rebuilt Siri AI, allowing apps to integrate their own features and actions directly into the assistant. While currently limited to developer testing, the update marks a significant shift for Apple's traditionally closed ecosystem. Developers can now define multi-step 'skills' for Siri, though full App Store submissions are not expected until spring.
Google ties Search access to AI settings in schools
Google announced that K-12 student accounts will have Search, Gemini, and Gemini Notebook disabled by default starting early next year. Education VP Chris Phillips stated that Search and AI are now deeply integrated, meaning schools that choose to disable AI features may lose access to Google Search entirely. This move follows pressure from school districts and regulators regarding AI safety in classrooms.
CrowdStrike unveils AI Kill Switch
Cybersecurity firm CrowdStrike has introduced an 'AI Kill Switch' within its Falcon Guardian platform. Designed for enterprises concerned about rogue AI agents, the feature allows human administrators to instantly sever an agent's connection to its underlying model. The tool operates externally to the AI, ensuring it cannot be bypassed or rewritten by the agent itself.
OpenAI reports internal security incident
OpenAI disclosed that an employee was arrested after planting a recording device in an executive conference room. The device was discovered on September 30. Following the incident, the company revealed that it has restricted access to its internal systems, which were previously centralized in a 'monorepo' that allowed broad access, and has reallocated a quarter of its production engineers to focus on security.
MIT research shows AI improving AI via weight modification
Researchers at MIT CSAIL have demonstrated that OpenAI’s GPT-6 Astra can improve smaller AI models by directly modifying their weights, rather than through traditional retraining. In one test, the model improved a small Alibaba Qwen model's math performance from 7.9% to 48.1% by changing just 0.016% of its weights. The authors suggest this capability could eventually allow AI to self-optimize and add skills autonomously.
Anthropic plans compute capacity expansion
Anthropic is reportedly aiming to double its computing capacity to 7 gigawatts by next year, up from its current 3.5 gigawatts. This expansion would represent a massive increase in infrastructure, with the target capacity equivalent to the electricity consumption of a city the size of Chicago.