The technology industry has developed a peculiar habit: rolling out powerful new features and burying the off switch in settings menus three layers deep. Generative AI has become the latest offender in this pattern of dark patterns and consent theater, where users are nudged toward participation rather than presented with genuine choice.
The Opt-Out Illusion
Every major tech platform has adopted the same playbook with AI features. Gmail's writing assistance. Microsoft's Copilot integration. Apple's on-device processing features. Google's AI Overviews. They all ship enabled by default, forcing millions of users into an implicit opt-in through inaction. The burden falls on users to discover these features exist, navigate labyrinthine settings menus, and actively disable them—often without clear guidance on what exactly they're opting out of.
This isn't accidental design. It's a calculated business decision dressed up as innovation. Default settings are sticky; most users never change them. Companies know this. They've known it for decades. Yet the industry continues to treat opt-out as a reasonable compromise between user autonomy and product adoption.
What's Actually at Stake
The stakes with AI are qualitatively different from previous feature rollouts. When a search algorithm changes, users might get different results. When a social media feed algorithm shifts, the content mix changes. These are important issues, but they pale in comparison to what generative AI systems can do.
AI systems trained on user data, integrated into email, search, documents, and communication tools, fundamentally change how our information is processed. They introduce new security vectors, privacy implications, and computational costs. They can hallucinate confidential information. They can perpetuate biases at scale. They represent a departure from previous computing paradigms—one that warrants explicit user consent, not reluctant acceptance through negligence.
For professionals handling sensitive information—lawyers, doctors, financial advisors, journalists—these features pose genuine risks. Yet they're enabled by default, requiring users to become security experts just to maintain their current threat model.
The Regulatory Momentum
Privacy regulators are beginning to take notice. The European Union's approach to AI regulation, embodied in emerging frameworks, increasingly emphasizes that users should have meaningful control over how their data interacts with AI systems. The FTC has begun scrutinizing dark patterns and deceptive default settings more aggressively. State-level privacy laws are catching up to the reality that consent theater isn't actual consent.
Yet the industry continues as if these signals don't exist. Each new AI feature rollout repeats the same cycle: enabled by default, buried in settings, defended with euphemistic language about user choice and convenience.
Why Opt-In Actually Makes Business Sense
The counterargument—that opt-in discourages adoption—deserves scrutiny. Yes, opt-in features see lower adoption rates than opt-out defaults. But that's not inherently a problem. It's actually a feature, not a bug. Features that users actively choose tend to be used more intentionally. They generate more genuine engagement. They avoid the resentment that comes from discovering you've been automatically enrolled in something unfamiliar.
Companies that have moved to opt-in models for sensitive features haven't collapsed. The feature adoption numbers drop, sure, but user satisfaction and trust typically increase. In an era where tech companies are facing erosion of user trust, that's not a negligible trade-off.
Moving Forward
The solution isn't complicated: sensitive features—especially those involving data processing, computational costs, or privacy implications—should require explicit, informed user consent before activation. This should be the default expectation, not a surprising exception.
Companies can still promote these features aggressively after opting in. They can highlight benefits in onboarding flows. They can offer guided tours. They can make adoption as frictionless as possible within the bounds of genuine consent. But the initial burden should fall on the company to persuade users, not on users to discover they've been enrolled in something they didn't choose.
The future of user trust in AI depends on this shift. Opt-out might be convenient for product teams. But opt-in is the only approach that respects user autonomy while still allowing genuine innovation to flourish.
