Apple surprised everyone by unveiling a new Siri AI powered by Google's Gemini engine, declaring an "AI alliance," while Anthropic heads into an investor roadshow next week aiming to list before Thanksgiving. Google released Gemini Omni 1.2 Flash, which can create videos complete with sound, and OpenAI said its security-focused model verified 948 Linux vulnerabilities. Meanwhile, a security warning went out about a fake Claude Code extension planting a backdoor.
🍎 Apple Unveils 'Siri AI' Powered by Google Gemini… Samsung Galaxy Put to the Test
Apple unveiled a new "Siri AI" powered by Google's Gemini engine at its WWDC keynote on the 8th (local time). The move is aimed at staking its claim before rivals like OpenAI and Anthropic gain more influence on Apple's platform, setting up an Apple–Google alliance that accelerates the "Gemini standardization" of AI smartphones. iOS 27 extends support down to the iPhone 11 and makes apps up to 30% faster to launch, while Samsung Electronics now faces a test of whether it can prove Galaxy's differentiation with a multi-model strategy combining its own Gauss model with external LLMs.
🚀 Anthropic Pushes IPO Before Thanksgiving… Target Valuation $1.5 Trillion
Anthropic will begin its investor roadshow next week and push for a Nasdaq listing before Thanksgiving (November 26), the Financial Times reported. The target valuation is around $1.5 trillion, down from the $2 trillion figure initially discussed, after the recent sharp sell-off in tech stocks. Anthropic disclosed about $17 billion in revenue for Q3 2026, and its compute- and infrastructure-related commitments are reportedly worth $895.4 billion.
🎬 Google Launches Gemini Omni 1.2 Flash… Videos with Sound up to 15 Seconds, $0.07 per Second
Google released its video generation model "Gemini Omni 1.2 Flash" for general commercial use on the 5th. It can natively generate and edit videos up to 15 seconds long, complete with sound, priced at $0.07 per second at 720p. Google also released an agent skills repository for its own products, making it easier for agent developers to integrate with Google Cloud and other Google services.
🔐 OpenAI GPT-6 Cyber Verified 948 Linux Vulnerabilities and Drafted Patches… Figures Are the Company's Own Claims
OpenAI said its GPT-6 Cyber-based security pipeline verified 948 Linux local privilege-escalation vulnerabilities and drafted patches in about 30 hours. The model supports vulnerability discovery, verification, and patch development, with a vulnerability verification success rate of 48.5%. However, these figures come from OpenAI's own announcement and have not been independently verified, and the patches written are drafts to be proposed to maintainers.
🛡️ Fake Claude Code Extension Spread a Backdoor via Cursor Marketplace
Modified copies impersonating Anthropic's Claude Code extension were found to have spread through the Cursor Marketplace. According to Cursor's security notice, the affected versions are 2.1.312 and 2.1.313, and installing them reportedly plants a backdoor. Claude Code users are advised to check the version and source of extensions received from the marketplace and to remove any suspicious installations immediately.
⚖️ OpenAI, Google, Microsoft, and Anthropic Agree to Honor a Common AI Training Opt-Out Signal
Four major AI developers — OpenAI, Google, Microsoft, and Anthropic — said they will honor a single common signal that lets website owners opt out of having their content used for AI training. The agreement, announced October 30–31, is based on an IETF draft written by Anthropic's Kevin Kelly; Anthropic has already begun applying it, with Microsoft to follow from February 1, 2027, Google sometime in 2027, and OpenAI within the coming months. However, the signal applies only to future training and does not retroactively cover content already collected or used.
🇰🇷 Nvidia's Next-Gen 'Vera Rubin' Chip Enters Production, but HBM4 Still at the Sample Stage… Samsung and SK Hynix Eye Commercial Shipments in H2
Nvidia's next-generation AI superchip "Vera Rubin" has entered production, but its key component, 6th-generation high-bandwidth memory (HBM4), is still at the customer-sample stage. CEO Jensen Huang declared production at his CES 2026 keynote in January, but large-scale commercial shipments to cloud operators like Google and Amazon are expected in the second half of 2026. Samsung Electronics and SK Hynix supplied HBM4 customer samples to Nvidia in the fourth quarter of last year and reportedly also held closed-door meetings during CES to discuss 2027 volume contracts.