Bypassing Safari 17's advanced audio fingerprinting protection
A blog post from a commercial fingerprinting provider shows how Safari 17’s new “advanced” audio anti-fingerprinting feature can be bypassed by averaging out the random noise Apple adds to Web Audio results. Commenters argue this highlights both the technical difficulty of preventing browser fingerprinting and the broader power imbalance between users and companies that track them, especially when such tracking is justified as fraud prevention. Many call for stricter legal limits on fingerprinting, more granular browser controls over powerful APIs like audio and GPU, or privacy-preserving, per-site identifiers that can’t be used for cross-site tracking.
Purpose and Ethics of Fingerprinting
- Many object that a company selling fingerprinting tech is publishing a “how to bypass privacy” guide while claiming it’s for fraud detection.
- Strong concern that user preference not to be tracked is overridden in the name of fraud prevention, likened to justifying wiretapping or attacks on encryption.
- Others argue some fingerprinting is genuinely useful for payments, abuse prevention, and fraud, but concede it is easily repurposed for advertising and profiling.
- Several see the vendor’s demos and pricing as oriented toward marketing/personalization, with “fraud only” language viewed as cover or CYA.
Technical Discussion: Audio Fingerprinting & Safari 17
- Safari’s protection adds random, uniform noise to Web Audio output.
- Commenters note this is weak: with enough samples, an attacker can average out the noise and recover the underlying deterministic differences.
- Differences between browsers arise from floating‑point behavior, implementation details, SIMD code paths, OS libraries, and anti-aliasing choices in Web Audio oscillators.
- Suggestions include: per-origin or time-rotating deterministic noise keyed by a secret; or fully standardizing deterministic audio algorithms across browsers.
Broader Web API and Browser Design Concerns
- Strong skepticism about exposing rich audio, GPU, and similar APIs to every website; many see this as driven by Chrome/Google and app-like web ambitions.
- Some want such APIs behind explicit permissions, or an opt‑in “rich JS engine,” keeping normal browsing to basic DOM/XHR.
- Others counter that permissions prompts are annoying, widely ignored, and themselves degrade UX.
Other Fingerprinting Vectors & Defenses
- Audio fingerprinting alone mostly narrows to browser/OS/hardware class but can be combined with IP, screen, timing, WebGL/GPU, etc. to become highly identifying.
- GPU/WebGL fingerprinting (e.g., DrawnApart) is cited as another potent vector.
- Disabling Web Audio/WebGL/JS or using unusual settings may itself make a user more unique. Tor and Mullvad browsers aim for a single shared fingerprint, but effectiveness depends on adoption.
Law, Policy, and Future Directions
- Some want fingerprinting, at least for advertising, made illegal; others note it is hard to define in law and that regulators may allow it for fraud/abuse/CP detection.
- Multiple comments argue privacy needs stronger legal codification because technical countermeasures alone are an endless, losing arms race.