California Deepfake Law: Governor Newsom's Enacted Legislation Explained

A head of state appears on video, alive and well, sipping coffee at a Jerusalem café. Within hours, Elon Musk's AI chatbot Grok flags the footage as a deepfake. The café owner scrambles to release corroborating photos. The Prime Minister of Israel is forced to post additional videos just to prove he still exists. If that scenario doesn't make you rethink how you handle video evidence in your own work, nothing will.

The deepfake panic in the headlines is mostly framed as a politics-and-celebrity problem. It isn't. It's an evidence problem. And if you're an investigator, attorney, or fraud analyst who relies on photo or video to make a case, the ground is shifting under you right now, whether you've noticed or not.

---

How Deepfake Laws Shape Legal Architecture

Here's the number that should stop you cold: as of January 2026, USA Herald reports that 47 U.S. states have enacted deepfake legislation, with 46 states addressing the creation or distribution of explicit deepfakes and 28 specifically targeting political deepfakes. That's not a fringe legal experiment. That's a near-universal legislative response happening faster than most industries have even clocked the problem. This article is part of a series, start with Deepfake Detection Accuracy Gap Investigator Workf.

South Dakota and Washington are among the most recent movers. Akin Gump's AI regulatory tracker documents South Dakota's SB 164, which mandates disclosure requirements for deepfake content in election materials, a disclosure-first approach that's survived First Amendment scrutiny, unlike California's outright prohibition law, which a federal judge blocked in 2024. Washington's governor signed complementary legislation targeting identity rights. The legislative trend is unmistakable: synthetic media is no longer a gray area. It has a legal address now.

California Deepfake Legislation and the Harassment Question

California deepfake legislation sits at an odd crossroads right now. The state passed some of the earliest deepfake bills in the country, aimed mostly at election-season manipulation and non-consensual explicit imagery, but a federal judge blocked the broadest election-related prohibition in 2024 on First Amendment grounds. That doesn't mean California deepfake law disappeared, it means the surviving provisions are narrower, more disclosure-focused, and more carefully targeted at specific harms like deepfake harassment and non-consensual imagery rather than blanket bans on synthetic political speech.

Harassment is where California deepfake law is on its firmest legal footing. Courts have historically given states much wider latitude to regulate targeted harassment of a real person than to regulate political speech generally, and California's remaining deepfake provisions lean into that distinction. If you're documenting a case involving synthetic imagery aimed at a specific individual, the harassment angle is often the strongest and most durable legal theory available under current California deepfake law.

Meanwhile, federal courts are quietly drafting their own answer. Proposed amendments to Federal Rule of Evidence 901, specifically a new Rule 901(c), would establish a two-step burden-shifting process for disputed digital evidence. As the University of Illinois Chicago Law Library explains it: challengers would first need to present evidence sufficient to support a finding of AI fabrication, and if they clear that bar, the burden flips. The proponent of the evidence would then need to demonstrate it's more likely than not authentic. That's a materially higher standard than what traditional chain-of-custody doctrine requires. And it's being written right now, for courts that will hear cases in the next two to three years.

State Deepfake Bills Beyond California

Other state legislatures have watched California's litigation carefully before drafting their own deepfake bills, which is part of why so many newer laws lean on disclosure requirements instead of outright bans. A disclosure-first structure is harder to challenge on First Amendment grounds than a straight prohibition, and lawmakers in nearly every state now drafting deepfake legislation know that. For investigators working across state lines, that means the same piece of synthetic video might be legal to distribute with a label in one state and flatly prohibited in another.

---

AI Deepfakes Made the "Video Equals Truth" Assumption Outdated

Modern deepfake generation systems, the same ones that had half the internet convinced Netanyahu was dead, can now replicate facial expressions, voice tone, and speech patterns with accuracy that defeats the naked eye. Not sometimes. Routinely. The University of Baltimore Law Review puts it plainly: deepfakes make it "difficult for courts to ascertain the authenticity of digital evidence," and "traditional methods will be challenged." That's law review language for: your old workflow is broken.

Consider what's already happening at the platform level. YouTube has expanded its deepfake detection tool specifically for journalists and public figures. Sony has flagged and removed more than 135,000 AI-generated deepfake songs from streaming services. Alethea has partnered with Reality Defender to embed deepfake detection directly into its Artemis platform. Zoom is integrating Pindrop's voice security to flag synthetic audio in enterprise calls. The platforms are building detection infrastructure at scale, which means the implicit message to everyone downstream is: you should be doing this too. Previously in this series: Political Deepfakes Video Evidence Authentication .

Deepfake Harassment, Privacy, and the Assembly Bill Pattern

Deepfake harassment cases often overlap with privacy claims, and that overlap shapes how California deepfake law gets applied in practice. A person whose likeness is used without consent in a fabricated video can sometimes bring both a harassment-based claim and a separate privacy claim, depending on how the assembly bill underlying the statute was written. Investigators should treat these as related but distinct legal theories, because the evidence needed to support a privacy claim isn't always identical to what's needed for harassment.

"Judges, not juries, should decide authenticity questions, with the court determining whether evidence is admissible and instructing the jury to accept it as authentic if approved." Professor Rebecca Delfino, proposal submitted to U.S. Courts — Federal Rules of Evidence Committee

That proposal from Professor Delfino is worth sitting with. She's not arguing that juries are stupid, she's arguing that deepfake authentication is a technical gatekeeping question, not a credibility question. The difference matters enormously for investigators. If judges start deciding authenticity before evidence ever reaches a jury, your documentation needs to survive judicial scrutiny before the trial even starts. That's a completely different evidentiary standard than most investigators are currently building toward.

---

Two Urgent Shifts for Anyone Who Submits Evidence

The Jones Walker LLP AI Law Blog notes that litigators are now advised to address chain-of-custody questions during early litigation stagesnot at trial, and that courts may require disclosure of any AI-created or AI-manipulated materials during discovery. That's not future-tense caution. That's current best practice guidance from a major law firm telling its clients to get ahead of this now.

Here's the uncomfortable reality nobody's saying loudly enough: detection tools themselves are unreliable right now. No foolproof method currently exists to classify video, audio, or images as authentic or AI-generated, full stop. The Illinois State Bar Association is explicit about this: AI content detection technologies "have proven unreliable and biased." So the answer isn't to outsource your judgment to a single detection algorithm. The answer is layered methodology, metadata analysis, facial biometric comparison, collection documentation, and a written record showing exactly how you reached your conclusion. That's what survives a challenge. One tool that flags something as "real" does not. Up next: What 99 Percent Accurate Really Means Facial Recog.

This is precisely the context in which structured AI face comparison that produces documented, court-ready analysis becomes operationally essential, not as a nice-to-have, but as the difference between evidence that holds and evidence that gets picked apart before lunch.

---

Professionals Adapting to AI Deepfakes Gain Competitive Edge

Think about what the xAI lawsuits over alleged deepfake nude images of minors signal, plaintiffs' counsel in those cases are already building arguments around AI generation as a legal harm. That argument structure travels. Defense attorneys in unrelated criminal and civil cases are watching those lawsuits very carefully, picking up the vocabulary, understanding the technical arguments. The cross-examination question "How do you know this video wasn't AI-generated?" is coming to a courtroom near you regardless of what the case is about.

Federal and state rules are only going to get stricter from here. For investigators, that doesn't just mean more paperwork, it means a chance to stand out. The teams that can walk a judge through clear collection logs, biometric comparisons, and documented verification steps will be the ones whose evidence actually gets admitted and believed. Everyone else will be stuck arguing from the back foot while their video is treated as just another clip on the internet.

California's deepfake policy did not emerge from nowhere; it followed a string of viral incidents involving fabricated video of candidates and public figures during election season. Lawmakers drafted an early assembly bill specifically to address deepfakes appearing close to an election, reasoning that a fabricated video released days before voters go to the polls leaves little time for a rebuttal to circulate. That timing problem is still central to how California deepfake law treats election-related content differently from other categories of synthetic media.

Digital replica statutes are a related but separate strand of California law worth understanding alongside deepfake rules. A digital replica provision generally addresses the unauthorized use of a real person's voice or likeness in commercial or entertainment contexts, which is a different harm than election manipulation or harassment. Investigators sometimes conflate the two, but the digital replica framework tends to focus on consent and compensation, while deepfake harassment and election statutes focus on deception and reputational harm.

It is unlawful under several California statutes to distribute certain categories of deepfake content without clear disclosure, and fines can apply on top of any civil liability a plaintiff pursues separately. The penalty structure varies depending on whether the content falls under the election-related provisions, the explicit-imagery provisions, or the harassment-focused provisions, so investigators should identify which statute applies before assuming a flat fine schedule. California is taking a narrower, more surgical approach after the 2024 court ruling, rather than trying to rewrite a single sweeping law that covers every category of harm at once.

California laws addressing deepfake technology continue to evolve as new legislative sessions convene, and practitioners should expect amendments rather than a static rulebook. Deepfake laws address issues that move quickly, a statute drafted around today's generation tools may need revision once new technology changes what's technically possible. That ongoing drafting process is part of why disclosure-based approaches have proven more durable than outright bans across state legislatures generally.

For anyone building a case that touches an election, a privacy claim, or a specific person's likeness, the practical lesson is the same: identify which California statute actually governs the fact pattern before assembling evidence. A case built around a fabricated video of a candidate near an election calls for different documentation than a case built around a private person's fabricated intimate image. Treating all deepfakes as legally interchangeable is the fastest way to have a strong piece of evidence excluded on a technicality that had nothing to do with whether the video was fake.

Non-Consensual Intimate Imagery and State Laws

Non-consensual intimate imagery is the category where deepfake laws move fastest, because most jurisdictions have laws regulating this specific harm even where broader political-speech statutes are still being tested in court. California's provisions treat synthetic intimate content the same way they treat other non-consensual intimate imagery, meaning the fact that the image was AI-generated rather than photographed does not weaken the underlying claim. Intimate visual depictions created without consent, whether through a face swap or a fully synthetic generation tool, criminalizes the conduct in much the same way regardless of the exact production method used.

State laws addressing this harm have converged more than they have diverged, which is useful for investigators working cases that cross state lines. A state law in one jurisdiction may use different terminology than a state law next door, but the underlying elements, lack of consent, intimate content, intent to distribute or harass, tend to repeat. That consistency across state laws is part of why non-consensual intimate imagery claims have survived First Amendment challenges more reliably than election-related deepfake laws have.

Federal Law and the Push for a National Act

Federal law has moved more slowly than state legislatures on deepfakes generally, but that gap is narrowing as Congress considers a federal act aimed at non-consensual intimate imagery specifically. A federal act would not replace state laws so much as sit alongside them, giving prosecutors a federal charge to bring when conduct crosses state lines or when a state law does not clearly reach the specific facts of a case. Supporters of a federal act argue that ai-generated deepfakes targeting private individuals deserve a uniform baseline rather than a patchwork of forty-seven different state approaches.

Any federal act that reaches the floor for a vote will need to survive the same First Amendment scrutiny that sank California's broader prohibition law in 2024, which is why drafters have leaned toward narrow, harm-specific language rather than sweeping bans. A federal act modeled on the harassment and non-consensual imagery theories that have already held up in court is more likely to survive than one modeled on the blocked election provisions. Investigators should watch this space closely, because a federal act would change which agency has jurisdiction over certain deepfake cases and could shift where charges get filed.

Election Deepfake Rules and Political Deepfakes

An election deepfake raises different legal questions than a harassment or privacy case, because political deepfakes sit closer to core political speech, which courts protect more aggressively than most other categories of expression. That's exactly why California's election deepfake provision ran into trouble in 2024, while South Dakota's disclosure-based approach to political deepfakes has so far survived. Investigators documenting an election deepfake should expect the applicable law to focus on labeling and disclosure rather than an outright creation ban.

Political deepfakes released close to a vote raise the timing problem discussed earlier in this article, there's often no practical way to correct the record before ballots are cast. Legislatures drafting election deepfake rules after California's setback have generally concluded that requiring a clear disclosure label is more defensible than criminalizing the underlying creation of the content. For investigators, that means an election deepfake case is more likely to turn on whether disclosure was provided than on whether the video was fake in the first place.

Platform Liability and Deepfake Detection Tools

Platform liability questions sit at the intersection of state laws, federal law, and private company policy, and they remain some of the least settled ground in this entire area. Platforms are not always required to proactively police every piece of synthetic content that users upload, but platform liability can attach when a platform is on notice of specific unlawful content and fails to act. That distinction matters for investigators building a case, because pursuing platform liability directly is often harder than pursuing the individual who created or distributed the content.

Deepfake detection tools built by platforms, like the ones described earlier from YouTube, Alethea, and Zoom, do not by themselves resolve platform liability questions, since a detection tool flagging content is a technical step, not a legal admission. Platforms adopting a public detection policy can reduce their own liability exposure while also creating a paper trail investigators can potentially request during discovery. As more platforms formalize a written detection policy, expect platform liability litigation to focus increasingly on what a platform knew and when, rather than simply whether harmful content existed on its service.

Deepfakes are illegal under a growing patchwork of state laws, but "illegal" does not mean "automatically excluded as evidence", investigators still need documentation showing why a given piece of content violates a specific statute. California deepfake law, federal proposals like Rule 901(c), and platform policy are all converging on the same practical demand: show your work. Whether the question is a state law violation, a federal act's reach, or a platform's internal policy on detection, the burden keeps shifting toward whoever is making the claim to prove it, not the other way around.

Governor Newsom signed the enacted legislation that now anchors California deepfake law, and understanding that signing history helps investigators explain to a court why a given provision exists at all. The enacted legislation covering non-consensual imagery moved through the legislature separately from the election-focused bills, which is part of why one branch survived the 2024 court challenge while the other did not. When a case cites a specific statute, checking whether Governor Newsom signed that provision as part of the earlier or later wave of enacted legislation can clarify which legal standard actually applies.

Revenge porn statutes predate most deepfake-specific enacted legislation, and California's approach to synthetic revenge porn largely borrowed the elements already established in those older revenge porn cases. A person creating a fabricated intimate image of someone without consent faces exposure under both the newer deepfake provisions and the older revenge porn framework, depending on how the image was produced and distributed. Investigators handling a revenge porn allegation involving synthetic media should check whether the underlying conduct also satisfies the newer, deepfake-specific statute, since prosecutors sometimes charge under both.

Sexual abuse claims involving fabricated imagery raise proof questions that differ from a straightforward harassment case, because sexual abuse statutes often require showing intent and a specific type of harm rather than just non-consensual distribution. When a case involves both sexual abuse allegations and deepfake technology, the deepfake-specific statute can sometimes provide an easier path to a charge than trying to fit the conduct into an older sexual abuse framework that wasn't written with synthetic media in mind. Investigators should document both angles rather than assuming one statute automatically covers the other.

Some newer proposals would use undeclared bots as a related regulatory target, extending the disclosure logic behind deepfake labeling laws to automated accounts that spread synthetic content without identifying themselves as non-human. The idea is that a platform user who encounters content generated or amplified by an entity that use undeclared bots faces a similar deception problem to someone viewing an unlabeled deepfake video. That proposed extension has not moved as quickly as core deepfake statutes, but it reflects the same disclosure-first thinking that has already proven durable in court.

Manipulative use of a person's image or voice is the underlying harm that most deepfake statutes are trying to reach, regardless of whether the specific provision is framed around elections, harassment, or non-consensual imagery. Any manipulative use that deceives a viewer about what a real person actually said or did tends to trigger the same basic legal concerns, even when the applicable statute and penalty structure differ. Investigators evaluating a new case should ask first whether the content reflects manipulative use of someone's identity, then work backward to the statute that fits.

Invasion of privacy claims often run alongside deepfake harassment claims, since a fabricated video or image frequently exposes private details or contexts the depicted person never agreed to share. An invasion claim can succeed even in situations where a harassment claim might struggle, particularly if the fabricated content reveals something private rather than simply portraying the person doing or saying something false. Investigators building a case with both angles should document the privacy harm separately from the harassment harm, since a court may treat the invasion theory on its own terms.

Sexual deepfake content involving a real person's likeness, without that person's consent, sits squarely within the categories of non-consensual imagery that California's surviving statutes were built to address. Whether the sexual content was fully AI-generated or created through a face-swap on an existing image, the statute generally treats the underlying harm the same way, because the core problem is the absence of consent rather than the specific production technique. Investigators should collect whatever technical evidence is available showing how the sexual image was produced, since that detail can affect which specific statute applies.

Criminal exposure under California's deepfake statutes depends heavily on which category of harm the conduct falls into, since the election-related provisions, the harassment provisions, and the non-consensual imagery provisions each carry different criminal thresholds. A criminal charge tied to non-consensual intimate imagery has generally proven easier to sustain after 2024 than a criminal charge tied to election-related content, given the different First Amendment postures of each category. Investigators should flag early whether a case is likely to proceed as a criminal matter or a civil one, since the documentation standards for each path are not identical.