The question came from a backend engineer in the third row, and it quieted a room of forty people. We were deep into a DeFi literacy workshop I run in Frankfurt, walking through how anyone can verify a lending protocol's solvency on-chain, when he asked the obvious thing: "If transparency is so powerful, why is the electrical grid — the thing that actually keeps us alive — protected by contracts nobody outside the utility can read?"
I didn't have a clean answer. Two days later, a report crossed my feed: Sam Altman had been discussing grid security with power utilities, eyeing a role for AI in protecting critical infrastructure. The CEO of the company building the most power-hungry technology on Earth, offering to defend the grid that feeds it. Set those two moments side by side and you get the real story of 2026 — not a cybersecurity announcement, but a referendum on who we trust to run the systems we cannot afford to lose.
To understand why this matters, you need the background the headline buries. Altman's outreach is not a cold start. OpenAI partnered with Anduril on counter-drone systems in late 2024, launched an "OpenAI for Government" arm, signed a pilot worth roughly $200 million with the Department of Defense, and quietly embedded researchers across the national laboratories. A grid-security conversation is the logical next step in a deliberate pivot — from selling API tokens to becoming a vendor of national infrastructure. When I trained senior bankers at Deutsche Bank's digital assets desk in 2024, I watched that same pattern from the other side. Institutions adopt new technology not when the code is elegant, but when the relationship feels safe. Altman is running that play again, with power utilities standing in for portfolio managers.
Here is the part that should concern anyone who still believes in decentralization. The electrical grid is not a modern software system. It runs on OT and ICS environments — SCADA control loops, DNP3, Modbus, IEC 61850 — where data looks nothing like the logs an IT security team ingests. Real-time constraints are measured in milliseconds, not block times. And unlike a DeFi protocol, a false positive does not cost you gas; it can trip a breaker and darken a city. I spent 2025 leading a Human-Centric AI initiative in Frankfurt, arguing that code must reflect human values before it touches consequential decisions. A probabilistic model guarding zero-tolerance infrastructure is precisely the category of decision I was warning about.
The technology simply does not fit the terrain. A large language model is a probability engine. It is brilliant at pattern recognition and terrible at guaranteeing the absence of an event. IT security operations tolerate that mismatch — a missed alert is absorbed by a human analyst kept in the loop. Grid operations cannot. Automation bias makes it worse: when a model speaks with confidence, dispatch operators stop second-guessing it, and the human backstop quietly erodes. We have watched vendors solve problems at a scale the customer does not have for years. The data availability layer arms race promised every rollup its own dedicated DA — and 99% of them never generated enough data to justify the spend. Complexity sold as safety has a long, expensive track record in this industry, and the pattern is repeating with AI cybersecurity.
The buyer and the vendor would be the same person. OpenAI is one of the largest new electricity consumers in the United States; its Stargate ambitions commit to gigawatt-scale demand, a single campus drawing as much as a mid-sized city. That makes Altman simultaneously the customer, the regulator's subject, and now the would-be security provider. In any other industry, this triangle would trigger a hard conversation about conflict of interest. Here it is being framed as public service. The real lesson is not that AI can protect the grid; it is that the pressure AI puts on the grid is what created the demand. There is a genuine feedback loop, and it runs in a direction the press release never mentions.
Crypto already built the part nobody is asking about. The grid's deepest vulnerability is not a lack of intelligence — it is a lack of verifiability. Nobody outside the utility can inspect the logic that controls the flow of power. In DeFi, we solved the inspection problem by default: the code is public, the state is public, and anyone can recompute the outcome. Back in 2021, when the Ethereum fee mechanism confused an entire user base, I built a visual guide to fee burning that got shared across dozens of communities, because clarity was the actual need — not more sophistication. A verifiable control plane for critical infrastructure is a solvable problem, and it looks more like cryptographic integrity and auditable ledgers than like a chatbot with a security clearance. The answer to "who guards the grid" should be "anyone can check," not "trust our model."
The regulatory backdrop makes this stranger still. The Biden-era executive order on AI risk was rescinded in January 2025, replaced by an action plan oriented toward acceleration rather than restraint. Europe's AI Act still classifies critical-infrastructure AI as high-risk, demanding human oversight and technical documentation, but the momentum in Washington runs toward deployment first and guardrails later. Altman is pitching grid security into a regulatory vacuum that his own industry helped create.
Then there is the question no one has answered: who is liable when the model is wrong? If an AI system misjudges a signal and a substation goes dark, does responsibility sit with OpenAI, the utility, or the exhausted operator who deferred to the machine? Until that question has a legal answer, no honest utility buyer can move from pilot to production. The silence on liability is itself the loudest signal — it tells you this is a relationship being seeded, not a product being sold.
And here we hit the honest dilemma. Crypto is not exactly ready to answer the call either. Interoperability improved again with Dencun, yet moving assets between rollups still feels orders of magnitude clumsier than a single centralized-exchange withdrawal — a reminder that "technically possible" and "actually usable" live in different buildings. Community is the only chain that cannot be broken, but that community still has a UX problem it keeps postponing. So we end up with two flawed suitors: a centralized AI giant offering speed and opacity, and a decentralized ecosystem offering verifiability and friction. The grid, which tolerates neither opacity nor friction, gets to choose between them.
This is where the contrarian reading earns its keep. The instinct among crypto natives is to celebrate any story that exposes Big Tech's centralization and assume the decentralized alternative wins by default. That is lazy. The grid will not adopt a public blockchain because it is philosophically appealing; it will adopt whatever survives a federal audit and a NERC CIP review. If this industry wants to be taken seriously here, it needs to stop pitching ideology and start shipping the boring, certified, human-legible infrastructure utilities actually procure. Community is the only chain that cannot be broken — but only after it passes compliance.
The competitive map sharpens the stakes. Microsoft holds roughly 49% of OpenAI's profit participation and also runs one of the world's largest security businesses — meaning the company's biggest backer is also its most entrenched rival. Anthropic chose the opposite path, delivering through trusted partners like Palantir and AWS GovCloud while keeping a deliberately low profile. Palantir brings government delivery muscle without owning a frontier model. CrowdStrike and Palo Alto own the real security telemetry. OpenAI enters with the strongest brand, the most advanced models, and a trust deficit in the exact domain where trust is the product — no federal security certifications, no delivery track record, no clearance infrastructure. Willingness to discuss grid security is not the same as readiness to run it. The word "eyes" in the reporting tells you this is relationship-building, not delivery.
The final risk is the one nobody puts in a headline: regulatory capture. A company that wants to protect critical infrastructure, while also consuming it and answering to its regulators, has every incentive to shape the safety standards in its own favor. That is not conspiracy; it is structural gravity. The grid asks a fair question. The uncomfortable answer is that the guard, the guest, and the lawmaker may be walking in through the same door — and until we can verify that separation, the safest system we can build is one where trust is optional, not demanded.
Few things are harder than watching a real threat — Volt Typhoon, Salt Typhoon — get used as the wedge to hand a single private actor visibility into the nation's most sensitive control systems. But community is the only chain that cannot be broken, and grid operators are a community too. They deserve tools they can audit, not models they must believe in.
The question that engineer asked still hangs in that Frankfurt room. Not long from now, someone on a utility board will ask it too — and whoever answers with verifiable proof rather than confident promises will define the next decade. Which side of that ledger do you want to be standing on?