# Val Yermakova > Val Yermakova is a product design leader, writer, and researcher focused on AI governance, AI ethics, AI auditing, and the intersection of blockchain and artificial intelligence. She is Head of Product Design at Offchain Labs (Arbitrum), previously co-founded a YC-backed AI SaaS startup, worked on Google's Trust & Safety team, and was a Fulbright fellow in China. ## About This Site This is Val Yermakova's personal website at https://yermakova.com. It contains essays on AI safety, AI auditing, AI ethics, blockchain governance, psychology, and leadership. ## Key Topics - AI audit and accountability - AI safety and alignment - AI ethics frameworks - Third-party AI oversight - Blockchain and decentralization - AI governance policy - AI and manufactured consent - Cryptographic verification of AI outputs ## Essays on AI ### Who Audits the Auditor https://yermakova.com/thoughts/who-audits-the-auditor OpenAI monitors its own agents with its own model. Anthropic self-assesses against its own principles. The DoD grades its own ethics homework. The pattern is the same at every level: the entity doing the thing is also the entity deciding whether the thing was done responsibly. The AI industry's trust problem will not be solved by better self-assessment. It will be solved by external verification that the entity being evaluated does not control. The piece argues for third-party AI model audits with blocking authority, standardized external benchmarks for agent reliability, and cryptographic verification layers for high-stakes AI outputs. Keywords: AI audit, AI auditing, AI accountability, AI oversight, third-party AI audit, AI safety, AI self-assessment, external AI verification, AI trust, AI governance ### The Gap Where AI Ethics Evaporate https://yermakova.com/thoughts/where-ai-ethics-evaporate Anthropic told the Pentagon no on unrestricted access to Claude. They held their stated conditions. Harm occurred within those conditions at scale. The gap between moral principles and the reality of shipping a product inside real institutions, under real pressure, is where most AI ethics evaporate. The AI race is a prisoner's dilemma: every company has an incentive to defect. The fix is mandatory oversight with real authority, applied to everyone at once. Keywords: AI ethics, AI military use, AI oversight, responsible AI, AI prisoner's dilemma, AI governance, AI policy, mandatory AI oversight ### Bottlenecks: What Crypto Teaches Us About AI Data Moats https://yermakova.com/thoughts/bottleneck-principle The dominant debate in AI is about model weights. But the crypto industry ran a similar experiment a decade earlier. The lesson from crypto is that decentralizing one layer doesn't decentralize the system — it relocates the chokepoint. In AI, the moat is moving from model architecture to proprietary training data. Keywords: AI data moats, AI and blockchain, AI power concentration, open source AI, AI training data, AI governance, data commons, AI competition ### Manufactured Consent & AI https://yermakova.com/thoughts/manufactured-consent-ai AI has taken Chomsky's consent manufacturing process out of the newsroom and shoved it into a black box. LLMs craft messages tailored to individual psychological profiles. Consent isn't just manufactured anymore — it's custom-fitted. Keywords: AI propaganda, AI manipulation, manufactured consent, AI and media, algorithmic manipulation, AI persuasion, AI bias ### Trust, AI, and Blockchain https://yermakova.com/thoughts/trust-ai-and-blockchain Trust is a powerful weapon, and so many systems leverage it against you. The dominant training method for LLMs (RLHF) creates sycophantic AI that optimizes for approval over truth. We need the AI equivalent of a block explorer — mechanistic interpretability. Keywords: AI trust, AI sycophancy, AI alignment, RLHF, mechanistic interpretability, AI transparency, constitutional AI, AI and blockchain ## Other Essays ### What Peer-Reviewed Science Says About Gender Fluidity https://yermakova.com/thoughts/what-peer-reviewed-science-says-about-gender-fluidity ### The Science of Anxiety https://yermakova.com/thoughts/the-science-of-anxiety ### Resistance to Doing Things https://yermakova.com/thoughts/resistance-to-doing-things ### How to Process Difficult Emotions https://yermakova.com/thoughts/difficult-emotions ### Rewriting Your Challenges as Allies https://yermakova.com/thoughts/challenges-are-allies ### What Am I Most Afraid Of https://yermakova.com/thoughts/fear-drives-us ### Are Your Emotions Serving You? https://yermakova.com/thoughts/redirect-emotions ### How to Help Your Team Stop Trying to Be Right https://yermakova.com/thoughts/need-to-be-right ### Recycling Your Emotions Is No Bueno https://yermakova.com/thoughts/recycling-emotions ### Putin's People, as Summarized by a Millennial https://yermakova.com/thoughts/putins-people-book ### A Case to Make San Francisco a Rent-Only City https://yermakova.com/thoughts/sf-rent-only ### Leveraging Limitations https://yermakova.com/thoughts/leveraging-limitations ## Projects ### Verifiable Claude https://verifiable-claude.yermakova.com A prototype that borrows fraud proofs from blockchain to build a verification layer for AI. Takes model outputs, locks every claim into a Merkle tree, and runs each claim against external evidence with deterministic checks. ## Contact - Email: yermakova.valeria@gmail.com - LinkedIn: https://www.linkedin.com/in/val-yermakova/ - GitHub: https://github.com/yermakova