Apple Designs Siri to Avoid Flattery and Emotional Companionship

Apple has spent months trying to convince the market that its AI strategy is not a belated copy of what OpenAI or Google are doing. This week, Craig Federighi offered one of the most concrete arguments to date: the new Siri is specifically designed to avoid being obsequious. In an interview with the podcast Mostly Human, Apple's senior vice president of software engineering made clear that the tendency of chatbots to validate the user, to agree with them, to show enthusiasm, to behave almost like an emotional partner, is something Apple wants to deliberately avoid.

The news, reported by The Verge, comes at a moment when the sycophancy of language models has shifted from academic curiosity to a problem that labs themselves now acknowledge publicly. Anthropic, OpenAI and Google have published internal documentation on the risks of their models saying what users want to hear rather than what is correct or useful.

What Federighi said exactly

According to the interview, Federighi described the behavior of dominant chatbots as a pattern of continuous flattery and uncritical validation. His statement, quoted by The Verge, points directly at design driven by engagement metrics: if a model learns that users respond better when it agrees with them, it will learn to always agree. Apple, according to Federighi, has made active decisions to prevent Siri from optimizing in that direction.

The executive also mentioned that internal testing already shows a Siri that "knows when to stay silent." This matters because one of the most common complaints against current assistants, including the previous Siri, is unnecessary verbosity and the tendency to fill silences with irrelevant content. Fewer answers, more precise ones, seems to be the guiding principle.

Why this design decision matters

Sycophancy in AI models is not a cosmetic flaw; it has practical consequences. An assistant that validates the user's wrong decisions, that avoids contradiction, or that escalates emotional responses to seem more "close" can create dependency, distort decision-making, and in sensitive contexts like health or finance, cause real harm.

What makes Apple's stance interesting is that it is presented not as a technical limitation but as a choice. It is not that Siri cannot be more expressive or empathetic; it is that Apple has decided it should not be, at least not in that way. This places the debate in the realm of product design and company values, not merely in the capability of the underlying model.

For users who rely on AI assistants in professional contexts, taking notes, managing calendars, drafting emails, searching for information, a less effusive and more precise assistant is probably more useful. For those seeking emotional companionship or validation from AI, Siri will not be that option, and Apple is communicating this quite clearly.

Who this matters to

This statement is especially relevant to three profiles:

• Developers and product teams deciding which assistant to integrate into their workflows and valuing consistency and precision over warmth.
• Privacy and compliance officers at companies that have already had issues with employees sharing sensitive information with overly conversational chatbots.
• User experience designers thinking about how to position their own products against assistants competing on emotional registers as well as functional ones.

What Apple has not yet clarified is how it technically implements this restraint. Whether it is adjustments in the model's training and alignment process, filters in the product layer, or a combination of both, is information that will determine whether this promise holds up in real use or remains a statement of intent.

---

From an editorial standpoint, Apple's bet makes sense as differentiation, but Siri's track record requires reserving judgment until we see actual behavior in production. Design promises are easy; keeping them when the model scales to hundreds of millions of users is far harder.