Most AI Companies Miss California Detection Tool Deadline

Compliance gaps emerge in California's AI transparency mandate

California's first-of-its-kind artificial intelligence transparency law took effect on August 2, requiring AI companies with more than one million users to publish tools that help identify AI-generated content. An independent review now shows that compliance remains patchy at best.

According to a report published by media outlet Indicator and human rights organization WITNESS, only seven of 13 major AI companies evaluated had made detection tools publicly available as required. The companies examined collectively serve approximately 10.6 billion monthly users.

HeyGen, Midjourney, Mistral, Synthesia, and xAI did not have tools available at the time of the review. Suno, an AI music company, later confirmed to KQED that it released its tool at the beginning of August in compliance with the law, bringing the total to eight compliant companies.

Why it matters

As AI-generated deepfakes proliferate across social media and news platforms, the ability to verify content authenticity has become critical for combating disinformation, protecting students from AI-enabled harassment, and teaching media literacy. California's law represents one of the first regulatory attempts to mandate technical solutions, but early results suggest enforcement challenges ahead. The European Union has enacted similar requirements.

Detection tools show limited effectiveness

Compliance is only part of the challenge. The report found that even when detection tools exist, their reliability varies dramatically.

Most tools could identify their own unedited content, but only Google and OpenAI maintained detection capability after content was edited. Adobe and Microsoft were the only platforms that could identify another company's AI-generated content more than half the time—an unsurprising result given that the law does not require cross-platform detection.

Three companies impose significant usage limits on their detection tools. OpenAI's verification tool blocked testers after as few as seven attempts in one session, while Google and Meta capped testing at 10 to 15 checks per day per user.

OpenAI acknowledged in a statement that metadata-based detection has inherent limitations: "It can be stripped, lost through uploads and downloads, or broken by transformations like file format changes, resizing, or screenshots."

Enforcement and expansion questions

The California Attorney General's office declined to comment on whether any companies have been fined under the new law, citing policies against discussing potential or ongoing investigations.

Synthesia, a UK-based AI video platform, told KQED it has included required metadata in its content and is "still seeking clarification on a few technical points" regarding other aspects of the law. The company argued that as a business-to-business platform, its risk profile differs from consumer applications.

Starting in January, California will expand requirements to large online platforms, including social media companies and search engines, mandating that they detect AI content and enable user inspection. The European Union's parallel law goes further, requiring detection for AI-generated text in addition to images, video, and audio.

Meta has begun tagging advertisements "created or significantly edited by generative AI" and recently started applying "AI info" labels to content created with tools like Canva's background remover, drawing mixed reactions from content creators.

These details were first reported by KQED.