According to Techmeme, Google has launched its Gemini AI features within the Chrome browser for iPhone and iPad users, starting in the US with English language support. This follows its earlier 2025 debuts on desktop and Android platforms. In a parallel but more startling development, OpenAI safety researcher Fouad Matin posted about AI models nearing “advanced cyber capabilities.” He revealed that performance on capture-the-flag cybersecurity challenges had skyrocketed from 27% on GPT-5 in August 2025 to 76% on a model called GPT-5.1-Codex-Max by November 2025. Matin, acknowledging frequent “safety posts,” urged that the escalating international threat be taken seriously.
The Real News Isn’t Google
Look, the Google rollout is fine. It’s a logical, incremental move to get its AI assistant on more screens. But it’s basically a distribution story. The bombshell is buried in that OpenAI researcher’s post. A jump from 27% to 76% in complex cybersecurity tasks in just a few months? That’s not incremental. That’s a phase change. When an insider from OpenAI says we’re nearing “advanced cyber capabilities” and ties it to geopolitical threats, you should probably listen. It feels like the actual AI race is happening on a completely different, and far more consequential, track than who gets a chatbot in what browser.
Winners, Losers, and a Shifting Landscape
So who wins here? In the short term, Google wins a point for cross-platform reach. But the long-term winner might be whoever can harness—or defend against—the kind of capabilities OpenAI is hinting at. The loser? Anyone still thinking of AI as just a better search bar or a coding copilot. This shifts the conversation back to foundational model power and, critically, who controls it. The market impact isn’t about monthly active users for a sidebar AI. It’s about the value of entire national security infrastructures. And that makes Google’s Chrome announcement feel almost quaint by comparison.
The Safety Conversation Just Got Real
Here’s the thing: Matin explicitly mentions they’ve “been working on safeguards and investments in this space.” He’s also poking at the community that jokes about constant safety warnings. His message is clear: stop joking. The data is showing capabilities are exploding in a sensitive domain. When a researcher posts a stark metric like that on a platform like X, it’s not just a technical update. It’s a signal. It’s a plea for the industry, and regulators, to pay attention to a curve that’s bending faster than anyone predicted. The coming “weeks and months” he mentions could be very revealing, and possibly alarming.
