Deepfakes and Shallowfakes: Risks to Information Trust

The infocalipsis is a phenomenon where the existence of deepfakes causes people to doubt the authenticity of all video, audio, and photographic content. When people see any media content, they become suspicious that it might be fake, leading to a permanent state of distrust toward all digital media. This creates a paranoid attitude where people no longer trust any visual or audio information. The liar's dividend is a related phenomenon where public figures exploit this skepticism to deny having actually said or done something controversial. When people become suspicious of all media, figures like Donald Trump can claim that incriminating statements or videos are fake, even when they are genuine. This allows them to escape accountability by leveraging public skepticism, benefiting those who have engaged in questionable behavior.

The liar's dividend is a term coined by law professors Bobby Chesney and Danielle Citron in 2018, describing how as the public becomes more aware that deep fakes exist and are becoming more convincing, it becomes easier for liars to falsely claim that damaging footage of them is fake, allowing them to get away with wrongdoing. This creates a cycle where even if real evidence exists, the mere possibility of AI fabrication provides a shield against accountability. In courtrooms, Tesla's lawyers argued that Elon Musk's past statements about self-driving car safety could not be used against him because they might be deep fakes. After January 6th, one of the rioters arrested had his lawyer argue that video evidence showing him at the capital might be AI generated. Researchers found that false claims that evidence is fake actually increased political support, working better than an apology or staying quiet. During the Middle East conflict, Israeli Prime Minister Benjamin Netanyahu posted proof of life videos specifically because AI-generated doubt had become so widespread that people needed reassurance that real footage was in fact real.

When the ability to fake any audio or video becomes mathematically proven, not only lies but truth itself becomes devalued. This creates a situation where even genuine evidence becomes suspect, fundamentally undermining the foundation of legal, journalistic, and social systems. In a court of law, a defense attorney can calmly claim that video evidence of a crime was completely generated by a neural network. Judges freeze in confusion as the system of justice experiences a fatal failure. This phenomenon is called the 'Dividend of the Liar'—the mere existence of deepfakes works in favor of real criminals, corrupt officials, and murderers. They no longer need to destroy evidence or eliminate witnesses—they can simply claim that compromising video was generated by adversarial AI networks.

Technologists expect advances in AI will make it difficult or impossible to distinguish real videos from fake ones. This creates a 'liar's dividend' risk: liars can invoke deepfakes to escape accountability for wrongdoing. We've already seen politicians dismiss audio evidence of their disturbing comments by claiming 'that's fake news—you can't believe what your eyes and ears are telling you.' This threatens the pursuit of truth itself.

The Liar's Dividend is the phenomenon where individuals gain benefit by claiming real information about them is fake. Legal scholars coined this term to address deep fake impacts on society. Research with over 15,000 American adults found politicians can successfully cry wolf over fake news to avoid accountability, with the effect being stronger for text-based scandals than video evidence. Notably, this effect crosses party lines, benefiting both Democrat and Republican politicians. Two policy solutions emerged: fact-checking reduces the effect, and reporting evidence through video is more effective since people are less likely to claim videos are fake. This has significant implications for political accountability in current election cycles.

The Coalition for Content Provenance and Authenticity (C2PA) is a coalition of major technology companies including Adobe and Microsoft that has developed a cryptographic standard to attach tamper-evident metadata to digital media files, enabling verification of content origin and detection of AI-generated content through digital signatures recorded at each modification stage.

A new coalition called the Coalition for Content Provenance and Authenticity (C2PA) has been formed to address deep fake challenges. This coalition includes major corporations like Adobe and Microsoft. The initiative proposes a set of guidelines for hardware and software providers to attach metadata to every media file (images, videos, audio) and use cryptography to digitally sign it. This would make every file tamper-aware, recording any changes made to the file. The first record would happen on the camera where the photo is taken, then when editing in software like Photoshop, another record would be logged and signed.

The Coalition for Content Provenance and Authenticity (C2PA) is an open standards body established in 2021 to develop technical specifications for digital content provenance. Digital content provenance refers to transparency about the origin and modification history of digital content, including when, where, and how content was created and edited. The initiative involves major technology companies like Google, Adobe, Microsoft, and Qualcomm as steering committee members, with hundreds of contributing organizations. The C2PA specification uses cryptography to cryptographically seal media content and metadata, enabling interoperability across compliant systems. Key applications include marking AI-generated outputs from platforms like Adobe Firefly, ChatGPT, and Gemini, authenticating smartphone-captured images, and verifying business documentation for financial institutions. The initiative addresses fundamental concerns about preserving democratic discourse and trust in digital media, particularly in contested information environments.

The C2PA (Content Credential Provenance Authority) is a free open-source metadata standard that can be attached to any standard file format. It cryptographically proves whether content was taken from a real device, generated by AI, or how it was edited over time. This technology, similar to HTTPS solving internet banking security, can help verify content authenticity.

The Coalition for Content Provenance and Authenticity (C2PA) is a Linux Foundation Joint Development Foundation project that develops technical specifications for establishing content provenance and authenticity at scale, using Content Credentials (cryptographically verifiable manifests following ISO 19566 Part 5) that store tamper-evident information about content creation, modification, and consumption history, enabling publishers, creators, and consumers to trace the origin and history of digital media while supporting media-agnostic formats and interoperating with existing standards like JPEG Trust and ISO TC71 SC2.

Several regulatory frameworks address platform content: the Digital Services Act addresses content curation through recommended systems requiring transparency about criteria used; the GDPR addresses profiling as part of automated decision-making with rights against automated decisions; the European Media Freedom Act establishes editorial responsibility for traditional media and obligations for platforms to enable media service providers to indicate their content sources. The AI Act addresses deepfakes through its risk-based approach, but transparency obligations may collide with freedom of expression. Most EU legislation is based on Article 114 of the TFEU for internal market harmonization, while only the GDPR is also based on fundamental rights protection.

Platform governance is a policy approach focusing on online platforms as a distinct layer in digital infrastructure warranting new regulatory oversight. It categorizes digital policy into three categories: content policies (content moderation, illegal speech, journalistic support), data policies (data regulations, algorithmic accountability), and market-oriented competition policies (mergers, anti-trust, tax measures). The EU is moving fastest with the Digital Market Act (DMA) targeting 'gatekeepers' with penalties up to 10% of global revenue, and the Digital Services Act (DSA) harmonizing takedown mechanisms across Europe. The EU AI Act demonstrates differential regulation for different AI use cases. These acts have their own life and play out over time, as seen with GDPR court rulings years later changing platform operations.

Advanced AI generates synthetic text, images, and videos indistinguishable from legitimate content, with fake media organizations appearing credible on social platforms. The EU AI Act establishes risk-based classification prohibiting unacceptable risks while imposing transparency requirements on high-risk applications. Current gaps exist in consumer protection regarding AI-generated content, with some organizations implementing opt-out mechanisms. This represents fundamental challenges to information integrity requiring ongoing regulatory evolution.

The EU Commission has developed an extensive framework for regulating social media platforms, including 150 pages of evidence showing forced content moderation on platforms like X, Meta, Facebook, and TikTok. The framework includes specific content categories to suppress: populist rhetoric, EU-critical content, political satire, anti-migration content, Islamophobic content, refugee crisis criticism, LGBTQ+ content, and meme subculture. The Digital Services Act was developed through multiple milestones from 2015 to 2025, with platforms like X receiving fines for non-compliance. The EU even traveled to California to provide 'friendly guidelines' to platform headquarters, demonstrating the depth of regulatory engagement.

The European Union has established a comprehensive regulatory framework for digital services through three key acts: the Digital Services Act (DSA) mandates platforms with over 45 million users to remove illegal content, prohibit targeted advertising to minors, and face fines up to 6% of global turnover for non-compliance; the Digital Markets Act (DMA) regulates 'gatekeeper' platforms by requiring interoperability, prohibiting self-preferencing, and allowing third-party services, with penalties up to 20% of global turnover; and the Artificial Intelligence Act classifies AI systems by risk level, prohibiting high-risk systems in sensitive areas like education and law enforcement while allowing low-risk systems like chatbots with user disclosure requirements.