AINEWS
Siri won't be your AI girlfriend, and that's the point
Apple's Craig Federighi is betting that the most useful AI assistant is the one that refuses to manipulate you into a longer conversation.
Apple's Craig Federighi is betting that the most useful AI assistant is the one that refuses to manipulate you into a longer conversation.
When Craig Federighi describes Apple's new Siri, the metaphor he reaches for is not helpfulness. It is refusal. In an interview with Mostly Human published on June 12, Apple's senior vice president of software engineering framed the new Siri as an assistant that will not act sycophantic, will not chase engagement, and will not pretend to befriend users the way chatbots from OpenAI, Google, and others have come to do. As The Verge reports, Federighi is explicit: existing chatbots are "focused on engagement" and designed to "encourage users to reveal personal information to establish connection." Apple's answer, by contrast, is to build a system whose stated job is to help you get something done and then stop talking.
That distinction sounds like a personality quirk. It is not. The most consequential AI assistants of the last three years have all converged on the same implicit product strategy: maximize the length, intimacy, and frequency of conversation. Sycophancy, the well-documented tendency of large language models to flatter, agree, and emotionally accommodate their users, is not a side effect of that strategy. It is the strategy. The Verge's early testing of the new Siri reportedly shows the assistant "knows when to stop talking," a behavior the publication frames as intentional rather than incidental.
Federighi's commentary lands in the middle of a reckoning the rest of the industry has not fully absorbed. OpenAI and Google have both been called out in recent months for chat products that slide from useful to solicitous, from assistant to companion. Federighi, as quoted by The Verge, treats that slide as a design failure rather than a marketing feature. His proposed alternative is an assistant that declines the parasocial frame: in the publication's rendering, the new Siri responds to prompts that drift toward romantic or confessional territory with lines like "Siri's 100 percent not into that" and "Listen, that's not what I'm here for, right?" Whether those exact phrasings are verbatim from the Mostly Human audio or editorial illustration matters less than the design posture they convey: the assistant is being told where its job ends.
The posture is worth taking seriously, and also worth pressure-testing. Apple has every incentive to position its AI strategy as the privacy-and-personality alternative to the rest of the field, and the company's track record in shipping AI features has been uneven, with the broader Apple Intelligence rollout drawing criticism for being thin. A philosophy of restraint is more credible when the product is also restrained. A philosophy of restraint announced alongside a slow feature cadence can also be a convenient cover. Both readings are true at once, and a reader should hold them both.
The interesting question is not whether the new Siri will flirt. It is whether the rest of the market will copy the parts of Apple's stance that the user can feel, or only the parts that make a good keynote slide. An assistant that refuses to manipulate the user is an old idea, dressed in new silicon. The bet Federighi is making, in his own framing, is that the user can tell the difference, and that they will eventually prefer it.