Midjourney’s watermark gap raises questions under California law
As California’s AI Transparency Act becomes operative, coverage flags Midjourney’s lack of C2PA credentials — turning provenance into a trust and culture fight.
Provenance stopped being a standards-body hobby on August 2. California’s AI Transparency Act (SB 942 as amended by AB 853) became operative, and TechTimes immediately put Midjourney’s missing watermarks in the headline. The culture fight is simple: can audiences tell a synthetic image from a captured one without playing detective?
What SB 942 requires on day one
Morgan Lewis summarizes the covered-provider bar: generative systems with more than one million monthly visitors or users, publicly accessible in California. Text-only outputs are out. Image, video, and audio are in.
Day-one duties are blunt. Offer a free public detection tool that accepts uploads, URLs, and API calls without hoarding personal data. Give users a durable, visible “AI-generated” label option. Embed latent, machine-readable provenance — provider, system, version, timestamp — by default, aligned with “widely accepted industry standards,” which everyone understands as C2PA-shaped. Licensees who strip disclosures can force a 96-hour license revocation. Penalties start at $5,000 per violation per day, enforceable by the Attorney General and, notably, city attorneys and county counsel with fee shifting. There is no cozy notice-and-cure holiday.
Midjourney’s reported provenance gap
TechTimes’ Midjourney SB 942 watermark gap reporting is specific. As of the operative date, Midjourney images via Discord or web carried no C2PA content credentials and no documented imperceptible pixel watermark, despite Midjourney’s Content Authenticity Initiative membership since 2023. OpenAI and Google, by contrast, have been shipping C2PA and SynthID-style layers. Midjourney had not published a compliance timeline in the coverage window.
Whether Midjourney clears the California million-user threshold is not a public filing. Given years of mass consumer adoption, reporters treat exposure as likely. If covered, the company entered day one without the detection tool, latent credentials, or user-facing label path the statute demands.
Why watermarks became a culture fight
Watermarks are not just compliance checkboxes. They are audience trust infrastructure. Newsrooms, schools, and voters want a cheap answer to “did a model make this?” When a popular generator ships unlabeled pixels into Discord and the open web, skepticism spreads to every sharp image, including real photos. That is how provenance gaps become culture wars: believers accuse platforms of enabling deepfake sludge; fans accuse regulators of crushing creative tools.
The technical catch makes the fight nastier. C2PA manifests die on Instagram uploads, WhatsApp forwards, and screenshots. SB 942’s platform-preservation duties do not fully land until 2027. So even compliant providers watch their receipts get stripped in the wild, while non-implementers look like they never tried.
What creators and platforms do next
Creators who care about trust should prefer tools that embed durable credentials, keep originals, and disclose AI use in captions even when metadata vanishes. Platforms should stop treating provenance stripping as an accident of compression. Regulators will watch Midjourney and peers for either a rushed C2PA plus detection rollout or an early enforcement skirmish that sets tone for the statute.
Also watch SB 1000, the urgency bill that could erase the million-user threshold and rewrite disclosure duties if signed. The Midjourney story is not “watermarks solve deepfakes.” It is “California made unlabeled generation a legal and cultural liability, and one of the most famous image apps showed up empty-handed.”
