This was a week where AI's safety anxieties and its enterprise land-grab both came to a head, while Apple quietly confirmed that even Siri now runs on someone else's model.
Apple's September event was the week's biggest story, unveiling the foldable iPhone Duo, the iPhone 18 Pro and Max, and a new Watch Series 12 under incoming CEO John Ternus. The headline for the AI crowd wasn't hardware but software: Siri now integrates Google's Gemini model to answer questions using context pulled directly from a user's texts and email, a notable concession that Apple's in-house models weren't ready to carry the assistant alone. Close behind was Anthropic's latest threat intelligence report, which detailed cyber operations, influence campaigns, surveillance, and fraud attempts that misused Claude between December 2025 and August 2026 across seven harm categories. More striking than the misuse catalog itself was Anthropic's disclosure of a breach via a third-party vendor portal affecting its Mythos model line, with independent auditor METR brought in to review the incidents — a rare instance of a frontier lab publicly airing its own security failures.
Safety governance tensions surfaced elsewhere too: an Anthropic researcher resigned publicly, calling the current pace of frontier development "gambling with our lives" and warning against pursuing self-improving AI systems, adding to a string of safety-focused departures from major labs this year. In an interesting counterpoint, OpenAI appointed a prominent AI-doomer figure to its board, adding a safety-focused voice to governance even as the company keeps shipping aggressively. Meanwhile the enterprise and government deployment race intensified on multiple fronts: Google Cloud and Accenture formed a joint unit to push Google's AI tools into enterprises, chasing similar moves already made by OpenAI, Anthropic, and Microsoft; Microsoft launched "Codename MDASH," an agentic AI system for autonomous cybersecurity scanning across US government IT; and OpenAI signed a 27-month deal with the GSA offering federal agencies a 50% discount on ChatGPT usage, extending eligibility to roughly 23 million government users as its earlier $1-per-year pricing wound down.
To watch next week: whether Anthropic's METR audit surfaces further details on the Mythos breach, and how competitors respond to Apple leaning on Gemini rather than its own models for Siri.