From new AI models to security risks and powerful new chips, here are eight important AI developments from the past few days. 1. OpenAI discovers a hidden recording device OpenAI found a sophisticated recording device hidden inside one of its offices. The company used its AI models to investigate the device and identify an employee who had planted it days earlier. OpenAI referred the matter to the FBI, which made an arrest. The employee was terminated and OpenAI said it had no indications that its model weights, source code or customer data had been improperly accessed or extracted. 2. Google launches Gemini 4 Argon Google has launched Gemini 4 Argon, its flagship AI model for complex tasks such as coding, finance, legal work, science, and cybersecurity. But how does it compare with other leading models? Here are some results: ➤ Overall performance: Argon scored 68.90%, ahead of Claude Sonnet 5.5 at 67.04% and Claude Opus 5.5 at 66.97%. ➤ Finance: Argon ranked first among 78 models on the Finance Agent v2 benchmark, scoring 65.40%. ➤ Building apps with AI: Argon scored 91.91% on Vibe Code Bench, but Claude Sonnet 5.5 performed slightly better at 92.39%. Argon is strong across several tasks, but it doesn't beat every competing model in every test. Google's introductory API pricing is $2 per million input tokens and $10 per million output tokens. These rates are scheduled to double after November 30, 2026. 3. Meta introduces Muse Wonder models Meta has unveiled Muse Wonder and Muse Wonder Full, two models designed to handle more complex tasks. Potential uses include handling insurance claims, understanding rental agreements and building apps. 4. Anthropic warns about potential AI self replication Anthropic has warned that smaller AI models could soon become capable of copying themselves across internet connected devices without needing large data centres. In its experiments, a model with 9 billion parameters successfully ran on eight of nine hardware types tested Anthropic warns that smaller models could reach this level by the end of 2026 if current progress continues. 5. AMD introduces Helion for AI agents AMD has introduced Helion, an open design for the computing systems needed to run AI agents. It allows companies to customise their CPU and GPU configurations and includes a token router that directs requests to different AI models. 6. Microsoft announces its Maia 300 AI chip Microsoft has announced Maia 300, its next-generation AI chip designed to run models and generate responses more efficiently. Expanded deployments are planned from 2027. 7. UK and Canada strengthen AI safety cooperation The UK and Canada have renewed their AI security partnership. Their plans include sharing information about AI risks, coordinating model tests, developing technical standards and exchanging expertise between their AI safety institutes. 8. X bundles Grok, Cursor and AI tools into new subscription plans X has introduced a new subscription lineup combining its social platform with Grok AI, Grok Bot and Cursor. The plans include: ➤ Plus: $8/month ⥂More AI usage, creator tools, a verification checkmark and fewer ads on X. ➤ Premium: $30/month ⥂More AI usage, access to Grok Bot, Cursor Pro features, and X Premium benefits. ➤ Super: $100/month ⥂Higher AI usage limits across Grok and Cursor. ➤ Ultra: $200/month ⥂The highest usage allowance among the new plans. Existing subscribers can keep their current plans or switch when the new option becomes available in their account settings. That's all for now. Talk soon.