California AB 2655 — Defending Democracy from Deepfake Deception Act: AI Compliance Requirements

HISTORICAL — AB 2655 was struck down 20 Aug 2025 (Kohls v. Bonta, Section 230 preemption); there is no live obligation pending the Ninth Circuit appeal. As enacted, this was a PLATFORM duty, not a creator duty: large online platforms had to develop and implement procedures using state-of-the-art techniques to identify reported materially deceptive election content and label it, no later than 72 hours after a § 20515 report, with a label stating "This [image/audio/video] has been manipulated and is not authentic" plus a clickable explanation. Non-English content had to carry the label in that language as well as in English (§ 20517).

HISTORICAL — AB 2655 was struck down 20 Aug 2025 (Kohls v. Bonta, Section 230 preemption); there is no live obligation to build, operate, or test takedown infrastructure pending the Ninth Circuit appeal. As enacted: large online platforms (at least 1,000,000 California users during the preceding 12 months, § 20512) would have had to (1) provide an easily accessible reporting mechanism open to candidates, elected officials, elections officials, and California residents (§ 20515); (2) respond to each report within 36 hours, describing any action taken or not taken (§ 20515); and (3) remove reported materially deceptive content within 72 hours during the removal window — 120 days before a California election through election day, extended through the 60th day after the election for content about elections officials (§ 20513). A candidate posting manipulated media of themselves had a safe harbor if it carried the disclosure "This ___ has been manipulated" (§ 20513(d)).

HISTORICAL — AB 2655 was struck down 20 Aug 2025 (Kohls v. Bonta, Section 230 preemption); there is no live prohibition pending the Ninth Circuit appeal. As enacted, the Act targeted "materially deceptive content": audio or visual media digitally created or modified — including deepfakes and chatbot output — such that it would falsely appear to a reasonable person to be an authentic record of what it depicts (§ 20512), where it falsely portrayed a candidate in a way harmful to reputation or electoral prospects, an elections official in connection with their duties in a way undermining election confidence, or an elected official in a way influencing a California election (§ 20513). The Act's remove/label duties fell on large online platforms — it did not prohibit individual creators from producing such content (creator-side rules were in separate statutes). Satire and parody were expressly exempt, as were qualifying periodicals and broadcast newscasts carrying the prescribed acknowledgments (§ 20519).