C2PA Cameras Do Not Survive Contact with Reality

Original Article ↗ Hacker News Discussion ↗

Efforts to cryptographically prove that photos and videos are “real” using C2PA-style camera signing are widely criticized as technically fragile and easy to bypass, especially on rooted or compromised devices and via analog workarounds like photographing a screen. Commenters worry such systems will create false confidence in “verified” media, be weaponized by powerful actors, and justify tighter hardware/software lock-in, while offering little protection against high-stakes disinformation. Some still see limited value in adding friction for low-effort fakes or for controlled workflows (e.g., newsrooms, compliance and licensing), but most argue that image provenance cannot reliably solve the broader trust problem.

Scope and Goals of C2PA

  • Many commenters argue C2PA can never provide “this is a real-world event” guarantees, only provenance/attestation for some chain of tools or publishers.
  • Some see it mainly as a way for organizations (newsrooms, agencies, advertisers) to show audit trails, not to prove ground-truth reality.
  • There’s confusion between marketing claims (“captured with a camera” badges) and what the tech can actually assure.

Technical Limitations and Attacks

  • Rooted or exploited phones can forge C2PA metadata; current Android approaches compared to “client-side password verification.”
  • Hardware attestation is viewed as stronger than software, but still vulnerable to exploits, supply-chain issues, and physical attacks (e.g., glitching, tampering with camera modules, breaking TPM-like components).
  • Several note that making the signing chip closer to the sensor helps but can’t be made perfectly secure, especially against well-resourced actors.

Analog Hole and “Real vs AI” Ambiguity

  • A recurring point: anyone can photograph a screen or a print of an AI image; this “air-gapping” attack fundamentally undermines any “real vs AI” claim.
  • Ideas like embedding LiDAR/depth data or sensor signatures are discussed; some think these might raise the bar, others think they’re still ultimately bypassable.
  • Commenters note even film negatives and darkroom work can be fabricated or derived from digital sources.

Usefulness vs Harm

  • One camp: C2PA is like a lock—imperfect but adds friction, blocking casual abuse and “slop” while not stopping nation-states.
  • Opposing camp: because it’s weakest in high-stakes cases (disinformation, legal evidence), it risks giving false confidence and making serious fraud easier to sell. Comparisons are made to DRM or misleading “trust” indicators.
  • Some argue the best outcome is society learning that photos no longer inherently prove anything.

Alternative / Narrow Use Cases

  • Suggested productive uses:
    • Internal provenance for media organizations.
    • Compliance and insurance in advertising (proving no AI was used when contracts or laws require it).
    • Helping curate ML training data by labeling at least some “original” imagery.
    • Low-stakes legal or commercial disputes where modest assurance is sufficient.

Regulatory and Ecosystem Concerns

  • Fears that C2PA-style systems will justify cryptographically locked-down devices, marginalizing rooted phones and open-source OSes, and creating a small cartel of “trusted” vendors holding keys.
  • Some propose legal mandates for labeling AI-generated content and showing origin country on posts; others doubt enforcement practicality and worry about unintended consequences.