https://manjchenna.com/essays/ Field notes on human judgment, AI, the EU AI Act, and who sets the rules AI runs on. By Manj Chenna, founder of Sanctity, Amsterdam. en https://manjchenna.com/essays/human-in-command.html https://manjchenna.com/essays/human-in-command.html Regulators are converging on a phrase: human-in-command. It is better than human-in-the-loop, and it will be diluted within a year unless somebody pins it down. Thu, 20 Aug 2026 09:00:00 GMT https://manjchenna.com/essays/the-three-layers.html https://manjchenna.com/essays/the-three-layers.html Trust in AI is treated as one problem. It is three: whose values the system runs on, who decides when it should not decide, and what actually happened. Values, Tue, 18 Aug 2026 09:00:00 GMT https://manjchenna.com/essays/the-model-on-the-invoice.html https://manjchenna.com/essays/the-model-on-the-invoice.html Your vendor bills you for the frontier model. How do you know that is what answers? On silent substitution in the AI supply chain and how serving-identity verification catches it. Thu, 13 Aug 2026 12:00:00 GMT https://manjchenna.com/essays/your-ai-changed-last-night.html https://manjchenna.com/essays/your-ai-changed-last-night.html The model beneath your product is not the one you tested. Providers swap, tune, and reroute in silence. The case for an independent record of machine behavior. Thu, 13 Aug 2026 12:00:00 GMT https://manjchenna.com/essays/the-expertise-layer-for-ai-agents.html https://manjchenna.com/essays/the-expertise-layer-for-ai-agents.html Models have a knowledge layer. Agents need something it does not provide: a path that takes the few decisions a machine should not make alone, and puts them in Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/cost-of-missing-oversight-balance-sheet.html https://manjchenna.com/essays/cost-of-missing-oversight-balance-sheet.html Removing the human looks like a saving on the day you do it. The bill arrives later, on different pages, attributed to other causes. Here is where to look for i Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/a-crumple-zone-with-a-job-title.html https://manjchenna.com/essays/a-crumple-zone-with-a-job-title.html A car's crumple zone is engineered to take the impact so the passenger does not. Put a human in that role inside an AI system, a person designed to absorb the b Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/designing-the-meaningful-override.html https://manjchenna.com/essays/designing-the-meaningful-override.html An override is only real if the human can actually reach it, has what they need to use it, and the system honors the result. Miss any of the three and you have Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/trust-is-engineered-not-advertised.html https://manjchenna.com/essays/trust-is-engineered-not-advertised.html You cannot label your way to trust. The word "responsible" on a page is not a property of the system, it is a hope about it. Trust is something you build in and Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-human-sla.html https://manjchenna.com/essays/the-human-sla.html Software has long sold service levels: we will be up this much, respond this fast. Human oversight needs the same move. A human-in-the-loop SLA turns "a person Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/automation-bias-index.html https://manjchenna.com/essays/automation-bias-index.html Automation bias is not a new discovery. The tendency to lean on reliable automation and stop checking it has been documented for decades. What we lack is a way Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/a-decision-too-important-to-automate.html https://manjchenna.com/essays/a-decision-too-important-to-automate.html The question is not whether a machine can make the decision. Often it can. The question is whether it should, and some decisions belong to a person for reasons Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/dignity-as-a-layer.html https://manjchenna.com/essays/dignity-as-a-layer.html Dignity appears in almost every AI principles page and almost no actual systems. The difference between the two is whether dignity is a line the system cannot c Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-trust-tax.html https://manjchenna.com/essays/the-trust-tax.html Every decision a customer cannot appeal to a human carries a hidden surcharge. They do not send you an invoice. They pay you a little less attention, a little l Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/giving-the-governed-a-vote.html https://manjchenna.com/essays/giving-the-governed-a-vote.html A system can be accurate and still have no right to decide. Legitimacy is the missing word in most AI conversations: not is it correct, but on whose authority d Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/whose-values-should-ai-hold.html https://manjchenna.com/essays/whose-values-should-ai-hold.html There is no view from nowhere. Every model that helps decide anything carries values, picked by someone, somewhere, usually without being asked out loud. The on Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-ai-trust-deficit.html https://manjchenna.com/essays/the-ai-trust-deficit.html Say "responsible AI" to an experienced buyer now and watch the small flicker of doubt. The phrase has been used too often by systems that were not, and the disc Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/equal-say-on-ai-values.html https://manjchenna.com/essays/equal-say-on-ai-values.html Expertise earns extra weight on questions that have a right answer. Values are not that kind of question. Confusing the two is how a small group ends up quietly Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/provable-vs-performed-oversight.html https://manjchenna.com/essays/provable-vs-performed-oversight.html There is a real difference between proving a human was present and proving a human was in command. Most oversight, even the kind with a cryptographic receipt, p Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-regulator-is-not-your-strategy.html https://manjchenna.com/essays/the-regulator-is-not-your-strategy.html Laws like the EU AI Act are starting to require a human over high-risk AI. That is a floor, and a welcome one. It is not a strategy. Treating compliance as the Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/rubber-stamp-rate.html https://manjchenna.com/essays/rubber-stamp-rate.html Count the times a human was asked to approve a machine's decision and changed nothing. Divide by the times they were asked. The closer that fraction sits to eve Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/every-model-is-an-opinion.html https://manjchenna.com/essays/every-model-is-an-opinion.html We talk about models as if they report the world. They argue about it. Every dataset is a selection, every objective a preference, every guardrail a value. Neut Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-rubber-stamp-problem.html https://manjchenna.com/essays/the-rubber-stamp-problem.html A better model should make oversight easier. Instead it often makes it emptier. The more reliably the machine is right, the more reflexive the human's approval Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/capability-is-downstream.html https://manjchenna.com/essays/capability-is-downstream.html The models keep getting better, and that is the story everyone tells. It is the wrong story to obsess over. Capability is an input. The rules that decide what i Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-whose-half-of-alignment.html https://manjchenna.com/essays/the-whose-half-of-alignment.html Alignment research mostly asks how to make AI follow human values. It quietly assumes there is one set to follow. There is not, and the assumption hides the har Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/from-data-labeling-to-live-judgment.html https://manjchenna.com/essays/from-data-labeling-to-live-judgment.html For years, "human in the loop" meant a person labeling examples so a model could learn. That human mattered, and then left before the system ever made a real de Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/governance-is-a-product.html https://manjchenna.com/essays/governance-is-a-product.html Most AI governance lives as a PDF: written once, filed, and quietly contradicted by what the system actually does. Governance that works is not a document. It i Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/how-an-ai-agent-asks-a-human.html https://manjchenna.com/essays/how-an-ai-agent-asks-a-human.html Strip away the jargon and it is simple. At a decision it should not make alone, the agent stops, hands the question to a person with enough context to judge, wa Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/when-expertise-outweighs-consensus.html https://manjchenna.com/essays/when-expertise-outweighs-consensus.html Counting heads is the right way to settle some questions and the wrong way to settle others. The trick is knowing which is which, because using the wrong rule q Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/how-to-measure-human-oversight.html https://manjchenna.com/essays/how-to-measure-human-oversight.html You cannot manage what you have not named, and human oversight of AI has gone unnamed for too long. Here is the small set of numbers that, taken together, tell Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/value-pluralism-for-ai.html https://manjchenna.com/essays/value-pluralism-for-ai.html A single set of values, applied to everyone, is either bland enough to be useless or specific enough to be unfair to someone. The honest design stance is plural Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/override-validity.html https://manjchenna.com/essays/override-validity.html It is not enough that a human overrules the machine. The override has to be right. A reviewer who confidently overturns correct decisions is not oversight, just Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/accountability-coverage.html https://manjchenna.com/essays/accountability-coverage.html For how many of your AI system's consequential actions could you name the specific person answerable for it? If the honest answer is "not many," the oversight i Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-vantage-point.html https://manjchenna.com/essays/the-vantage-point.html I did not arrive at this work from one place or one discipline. The thread that runs through it, a wariness of any system that decides for people without answer Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-next-decade-of-humans-and-ai.html https://manjchenna.com/essays/the-next-decade-of-humans-and-ai.html I will make a falsifiable bet. The next ten years of AI will be decided less by how capable the models become and more by whether we build the unglamorous infra Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/refusal-as-a-feature.html https://manjchenna.com/essays/refusal-as-a-feature.html A system that will attempt anything is not powerful, it is unbounded. The decisions an AI deliberately will not make, the calls it hands to a person or declines Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/when-the-ai-agent-acted-alone.html https://manjchenna.com/essays/when-the-ai-agent-acted-alone.html A model that predicts can be wrong on a screen. An agent that acts can be wrong in the world: the message sent, the account closed, the order placed. The new fa Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-trust-layer-for-ai.html https://manjchenna.com/essays/the-trust-layer-for-ai.html Reliability engineering gave us uptime, a number you can check instead of a promise you have to take. AI needs the same move for trust: a layer that produces ev Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/what-trustworthy-ai-requires.html https://manjchenna.com/essays/what-trustworthy-ai-requires.html Trustworthy AI is one of the most used and least specified phrases in the field. It is worth asking what it would take to deserve the word, rather than just pri Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/building-ai-governance-in-amsterdam.html https://manjchenna.com/essays/building-ai-governance-in-amsterdam.html It can look like a handicap to build under the strictest AI law in the world. I think it is the opposite. The hardest environment to satisfy is the best place t Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/model-size-is-not-trust.html https://manjchenna.com/essays/model-size-is-not-trust.html A bigger model is more capable. It is not, by that fact, more trustworthy. Capability and trust are different axes, and confusing them lets a very smart system Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/a-constitution-for-ai.html https://manjchenna.com/essays/a-constitution-for-ai.html The values a model holds are real whether or not anyone wrote them down. A constitution, in the loose sense, is just the decision to write them down: to make th Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-accountable-human.html https://manjchenna.com/essays/the-accountable-human.html When an AI agent makes a consequential call, there should be a person, a specific one with a name, who is answerable for it and could have changed it. A queue i Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/a-portal-for-human-judgement.html https://manjchenna.com/essays/a-portal-for-human-judgement.html If an AI agent needs a person's judgment, where does it go to get it? Most systems have no answer, just an informal favor from whoever happens to be around. Nam Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/when-should-an-ai-agent-escalate-to-a-human.html https://manjchenna.com/essays/when-should-an-ai-agent-escalate-to-a-human.html Too much escalation and the agent is pointless and the humans are buried. Too little and it acts alone on calls it had no business making. The art is in the lin Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-ai-kill-switch.html https://manjchenna.com/essays/the-ai-kill-switch.html Every serious system needs a way to stop. The trouble is that the stop control is the one feature nobody exercises, so it tends to exist on the diagram and fail Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/defence-in-depth-for-ai-decisions.html https://manjchenna.com/essays/defence-in-depth-for-ai-decisions.html No serious security team trusts a single wall. They assume any one control will fail and layer the next behind it. Human oversight of AI deserves the same humil Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-right-to-a-human.html https://manjchenna.com/essays/the-right-to-a-human.html When a machine makes a decision that changes your life, you should be able to ask for a person. Not a different machine, not a form, a human who can look again Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/human-in-the-loop-for-ai-agents.html https://manjchenna.com/essays/human-in-the-loop-for-ai-agents.html The phrase is old and the meaning has shifted under it. For models that only predicted, a human in the loop was a labeler. For agents that act, it has to mean a Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/when-an-algorithm-decides-alone.html https://manjchenna.com/essays/when-an-algorithm-decides-alone.html The danger is rarely a dramatic robot. It is a quiet rule, set once, far upstream, that then decides at scale while the people who wrote it are nowhere near the Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/audit-trails-for-ai-decisions.html https://manjchenna.com/essays/audit-trails-for-ai-decisions.html Most systems log plenty. Almost none log the thing that matters after a bad decision: enough for an outsider to reconstruct what happened, why, and who was answ Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-cost-of-no-human-in-the-loop.html https://manjchenna.com/essays/the-cost-of-no-human-in-the-loop.html The bill for missing oversight does not arrive as a fine. It arrives quietly, as liability you did not price, work you have to redo, customers who stop trusting Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/recall-for-a-decision-system.html https://manjchenna.com/essays/recall-for-a-decision-system.html A faulty car gets recalled. A faulty toaster gets recalled. A faulty model that made thousands of wrong decisions usually gets a quiet update, while the decisio Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-thought-economy.html https://manjchenna.com/essays/the-thought-economy.html When machines can do the work and recall the knowledge, the scarce thing left is judgment: deciding what matters, what is acceptable, and who answers for it. Th Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/escalation-fit.html https://manjchenna.com/essays/escalation-fit.html It is not enough that an AI system escalates to a human. It has to escalate the right things. Send too little and the agent decides hard cases alone; send too m Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/how-to-build-oversight-that-holds.html https://manjchenna.com/essays/how-to-build-oversight-that-holds.html Oversight that holds is not a warning label or a person cc'd on the outcome. It is a handful of design choices, each boring on its own, that together keep a hum Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/auditability-is-the-new-uptime.html https://manjchenna.com/essays/auditability-is-the-new-uptime.html A generation ago, software earned trust by publishing uptime, a number you could check instead of a promise you had to take. AI decisions need the same move, an Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-values-layer-for-ai.html https://manjchenna.com/essays/the-values-layer-for-ai.html Every model applies values. The values layer is the part of the system where those values are written down where you can see them, argue with them, and change t Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-cost-of-a-wrong-no.html https://manjchenna.com/essays/the-cost-of-a-wrong-no.html Companies watch the wrong yes, the fraud that got through, the loan that defaulted. The wrong no is quieter and often costlier: the good customer denied, with n Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/human-response-time-metrics.html https://manjchenna.com/essays/human-response-time-metrics.html Reliability teams never settled for one timing number, and neither should oversight. How long until a human is reached, how long until they respond, and how lon Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/time-to-human.html https://manjchenna.com/essays/time-to-human.html Operations teams already live by one number: how long until a problem reaches a person who can fix it. Point that same number at AI, and you can finally measure Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/accountability-inversion.html https://manjchenna.com/essays/accountability-inversion.html When a person and an AI share a decision, the authority drains to the machine and the blame stays with the human. The quiet defect under most AI oversight. Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-confirm-button-is-not-oversight.html https://manjchenna.com/essays/the-confirm-button-is-not-oversight.html A button that is only ever pressed measures nothing. If the human approves every time and changes nothing, that is a stamp, not oversight. Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/automation-bias-explained.html https://manjchenna.com/essays/automation-bias-explained.html The better an automated system gets, the less the human checks it. That is not laziness, and it is why the rare error slips through. Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/who-sets-the-rules-ai-runs-on.html https://manjchenna.com/essays/who-sets-the-rules-ai-runs-on.html The contest that matters is not how capable AI gets. It is who writes the rules underneath, and whether you are allowed to see them. Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/the-oversight-illusion.html https://manjchenna.com/essays/the-oversight-illusion.html A power nobody uses looks exactly like a power that was never there. That is why so much AI oversight passes inspection. Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/contestation-latency.html https://manjchenna.com/essays/contestation-latency.html How much time and context a human actually has to disagree with the machine. Three seconds per case is a turnstile, not a judgment. Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/oversight-budget.html https://manjchenna.com/essays/oversight-budget.html Human attention is fixed; the decisions are not. Divide one by the other and, past a point, each decision gets none. Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/liability-laundering.html https://manjchenna.com/essays/liability-laundering.html Some human sign-offs do not add judgment to a decision. They add a name to blame. Tue, 24 Jun 2026 12:00:00 GMT https://manjchenna.com/essays/how-i-got-here.html https://manjchenna.com/essays/how-i-got-here.html A nine-year-old in a 1990s computer lab, the two games that marked me, and the one idea about rules I have been circling ever since. Mon, 22 Jun 2026 10:00:00 GMT https://manjchenna.com/essays/why-my-logo-is-a-glider.html https://manjchenna.com/essays/why-my-logo-is-a-glider.html A small thing, explained once. Five squares from Conway's Game of Life, flipped into a question mark, and why hacking means growth, not theft. Mon, 22 Jun 2026 09:00:00 GMT https://manjchenna.com/essays/human-oversight-is-theater.html https://manjchenna.com/essays/human-oversight-is-theater.html The EU AI Act made human oversight the law. Almost no one is measuring whether it is real. The metric that would expose the difference, and the bet behind it. Sun, 21 Jun 2026 09:00:00 GMT https://manjchenna.com/essays/meaningful-override-rate.html https://manjchenna.com/essays/meaningful-override-rate.html An open standard for telling real human oversight of AI from theater. The definition, the method, and what makes a number worth citing. Sun, 21 Jun 2026 09:00:00 GMT https://manjchenna.com/essays/the-game-of-life.html https://manjchenna.com/essays/the-game-of-life.html Conway's Game of Life in plain English. Four simple rules, the patterns they grow, a board you can play, and what it has to do with who sets the rules AI runs on. Sun, 21 Jun 2026 09:00:00 GMT