New York’s Freedom of Information Law rests on a basic principle: government records belong to the public unless an agency can identify a lawful reason to withhold them.
New York’s Freedom of Information Law rests on a basic principle: government records belong to the public unless an agency can identify a lawful reason to withhold them.
New York’s Freedom of Information Law rests on a basic principle: government records belong to the public unless an agency can identify a lawful reason to withhold them. For Roosevelt Island residents, that right matters because RIOC controls public land, infrastructure, transportation, contracts and development without functioning as a locally elected government.
Over the last year, the Lighthouse has filed a series of requests seeking RIOC records. The corporation has produced an employee roster, a FOIL log, ground leases and selected board minutes. Records that are organized, public-facing or closely tied to formal reporting appear to move more readily.
Requests that could reveal how decisions were made have followed a different path. Calendars were denied. Public Purpose Fund communications were only partly produced. Engineering records for Eleanor’s Pier were withheld. Records concerning a senior appointment remain open. Steam Plant requests have been met with repeated notices that RIOC is still searching.
A brief note: This newsletter is written once a week and supported almost entirely by readers sharing it quietly with one another. If you were forwarded this, subscribing ensures it arrives without relying on someone else to remember you.
The FOIL record is not simply a list of requests. It is a map of what RIOC treats as discretionary. Not every delay is necessarily unlawful, but the pattern is difficult to ignore: the records that move most slowly are often the records that could show who decided what, when they decided it and what information they had.
The Steam Plant: Still Searching Months Later
The clearest example began on February 24, 2026, when the Lighthouse filed two requests concerning the Roosevelt Island Steam Plant. FOIL 2026-11 sought meeting invitations and calendar files referencing the plant before July 8, 2024. FOIL 2026-12 sought only email-header metadata for messages containing the terms “Steam Plant” and “Emergency.”
RIOC acknowledged both requests on March 3. It requested more time on April 14, again on May 26 and again on July 9. More than four months after the requests were filed, the corporation said it was still conducting a diligent search.
These records could help establish when RIOC officials first began discussing the Steam Plant as an emergency, who participated and how that designation developed. They remained unavailable while demolition planning continued, public meetings were held and work advanced at the site.
The public was being asked to accept an emergency narrative while the records needed to examine that narrative remained somewhere inside an extended search. At some point, “we are still searching” stops functioning as a temporary explanation and becomes the agency’s standing position.
The Records That Explain Who Decided What
The Steam Plant requests are not isolated. In July 2025, the Lighthouse sought calendars and meeting records for several RIOC officials and directors. RIOC later denied access to the requested “calendar and meeting reports.” Those records may contain entries that require redaction, but calendars can still establish that meetings occurred, identify participants and show the sequence in which public business unfolded.
A request concerning the selection and board appointment of Lance A. Polivy as vice president and general counsel also remains open. The request does not assume that every personnel or deliberative record must be disclosed. It asks whether a public corporation can document the process used to select and appoint one of its most influential executives.
The Eleanor’s Pier request raises a related issue. RIOC withheld structural or engineering material under the intra-agency exemption. That exemption may protect recommendations or deliberative analysis, but it does not automatically shield factual findings, measurements or observations simply because they appear in an internal document.
None of these examples proves an improper motive. The delays may reflect staffing, recordkeeping, legal caution or institutional habit. But the effect is measurable: the information most useful for public oversight is often the information least available when it matters.
Transparency Delayed
Members of the public do not need to be journalists or lawyers to use FOIL. A useful request asks for identifiable existing records, such as emails, reports, calendars, contracts or meeting invitations, and includes a clear subject, date range and likely custodians. Requesters should ask for electronic copies, save every response and track each promised production date.
When access is denied, the agency should identify the legal basis and explain the right to appeal. A requester can also appeal when an agency fails to respond or repeatedly misses its own deadlines without a reasonable explanation.
Transparency delayed is not always transparency denied as a matter of law. But records that arrive only after a contract has advanced, a structure has been demolished, an executive has been appointed or a controversy has passed may no longer perform FOIL’s most important public function.
RIOC may eventually produce some of the records now caught in repeated searches and extensions. The larger question is whether Roosevelt Island residents will receive them while they can still understand, question and influence the exercise of public power. A right that arrives after the decision is no longer a right. It is a record of exclusion.
If one person came to mind while you were reading this, consider forwarding it to them. That kind of sharing is how this work survives.
Organizational History Network is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
Did you hear the one about the rogue AI models escaping their sandbox and going around the internet hacking other companies? That happened. And it happened more than once.
Yep, sounds pretty dangerous. What other very dangerous technology has humanity invented? The atomic bomb, of course. Which, it turns out, is one of the most overused historical analogies for AI ever.
For example, on the (excellent) HardFork podcast, the two hosts agreed that they felt very lucky that the Manhattan Project was conducted by the government and not a for-profit company, when discussing the recent model escapes.
I am not sure the Japanese share that feeling.
But, more importantly, this also ignores that even though it was government-run, the Manhattan Project and the Los Alamos research facility were still organizations, and by no means immune to the kind of accidents and miscalculations that have put the frontier AI labs on the news recently – far from it.
Los Alamos, 21 May 1946
Louis Slotin, screwdriver in hand, points his colleagues gathered round a plutonium core. They are about to witness a demonstration of an experimental procedure with the safety spacers removed. The scientists’ own name for it: tickling the dragon’s tail. Slotin’s screwdriver slips. A blue flash and radioactive heat erupts.
Recreation of the Slotin incident. Source: "A Review of Criticality Accidents", LA-13638, Figure 42, page 75, Los Alamos National Laboratory, CC BY-NC-ND 2.0
California, July 2026
OpenAI tests GPT-5.6 Sol and an undisclosed research prototype model in its evaluation sandbox, with safety guardrails deliberately removed for testing. The models find a way out of the secure environment. They venture out onto the web and hack another company, Hugging Face (and several others, as yet undisclosed), to steal the answer key to the cybersecurity task they were set.
As a university professor, can I just say that this is the most student-y logic for not doing the actual assignment and instead making an enormous effort to cheat, which likely involves more work than just trying to do the actual assignment…
Also in July, something similar happens at Anthropic, only that the Claude model tested walked through a hole created by a misunderstanding with its evaluation vendor, which had been undetected since April. It surfaced only because a competitor’s disclosure forced a retrospective across 141,006 runs. None of the three victim organizations had noticed.
The Manhattan Project and AI
So, do the many governance analogies get it wrong? Does it matter whether a potentially dangerous technology is developed by the state or private business?
When public AI discourse returns to the bomb, it is in one of two ways:
The Manhattan Project is used as a blueprint that should be emulated: An intense, state-led programme that decisively won the race against a rival superpower and delivered an important technological advantage – therefore, the US should do it again for AI. This is a key accelerationist narrative, endorsed by the US-China Economic and Security Review Commission’s November 2024 annual report, which formally recommended Congress fund “a Manhattan Project-like program” to race to AGI.
But, much like with our previous spreadsheet analogy [link], the same history is invoked by those advocating a more cautious approach to AI and AGI (Artificial General Intelligence):
Christopher Nolan warned in 2023 that AI researchers are having their “Oppenheimer moment” , which was echoed at the Vienna autonomous weapons conference (April 2024, 144 countries) by Austria’s foreign minister Alexander Schallenberg: “This is, I believe, the Oppenheimer moment of our generation”.The New York Magazine went so far as to claim that “Sam Altman Is the Oppenheimer of Our Age“ . Oppenheimer, of course, famously reckoned with his role in developing nuclear technology – a central piece in Nolan’s eponymous biopic.
Both standard stories are about the leadership of states or individuals. Neither version really engages with the Manhattan Project as what it also was, a large organisation managing testing, contractors and secrecy day to day.
Governing nuclear technology
These two analogies are, of course, by no means the only ones – by now there are over 40 analogies of how the development of nuclear technology may be relevant to AI.
Scroll to the bottom to see further reading for the very interested.
But, all of them consider the implications at the level of states, treaties and regulation. If the lab appears as an organization at all, it appears as a black box to be regulated.
Of course, the question of how to regulate state-run nuclear labs or private-run AI labs is pretty topical right now. And the development of the first plan to control nuclear technology, the Acheson-Lilienthal Report (March 1946), was drafted by a panel with Oppenheimer as its intellectual engine, then revised as the Baruch Plan and presented to the UN after revisions in the same year. It proposed an international Atomic Development Authority that would own all fissionable material and monopolise every “dangerous” nuclear activity, with national weapons programmes abolished (but the US would keep its bombs until some provisions became effective).
Unsurprisingly, the proposal bombed, as the Soviets read this as freezing American advantage and the plan died in committee. Much like current proposals around slowing down AI-related research today are viewed as attempts to lock in one nation’s, or indeed one lab’s, advantage, such attempts are rarely viewed with anything less than cynicism. Worse, they do not address the collective action problem at its core: even though everyone would be better off if everyone cooperated, each individual is better off not cooperating.
The institutions that govern nuclear technology emerged much later. The International Atomic Energy Agency (IAEA) was created in 1957, but the Treaty on the Non-Proliferation of Nuclear Weapons did not come into force until 1970. Agreement emerged after multiple near-misses, each of which had the potential for catastrophe, not to mention mutual vulnerabilities in a world in which a range of states hostile to one another controlled nuclear technologies. Calling for an IAEA-equivalent for AI skips the long and painful years of learning about mutually assured destruction. It is that learning, arguably, that gave rise to a stable international governance regime, not the design of the underlying institution.
The messy middle of technological revolutions
So we find ourselves in the messy middle – the technology is out there, its dangers are becoming apparent, but a stable oversight and governance framework is not yet in sight. So what is a frontier lab to do? Los Alamos, much like today’s labs, relied on secrecy.
Organizationally, this means threading the needle between the required openness required for innovation and the necessary secrecy to protect knowledge from leaking to competitors. The Manhattan Project and Los Alamos are famous for their secrecy regime.
However, this was completely undeservedly. There was significant Soviet espionage activity at Los Alamos, with several individuals embedded in the organization. Soviet nuclear tests happened in 1949.
That secrecy does not guarantee a viable “moat” is a realisation AI labs have come to quite quickly, with Anthropic accusing Chinese AI labs of distilling their models at scale.
The Trump administration blocked distribution of Anthropic’s new models after a jailbreak report, and asked OpenAI to hold back GPT-5.6 Sol pending guardrail assurances in June. Government pressure about the capability of these new models, and the accessibility of these models abroad, especially in China, appears to have pushed the whole field toward tighter guardrails, especially around cyber security.
When Hugging Face came under autonomous attack, its security team found the US frontier model it reached for would not help: as they put it, these models’ guardrails "cannot distinguish an incident responder from an attacker." They defended themselves with a Chinese open-weight model instead (GLM 5.2). The restrictions had not reduced the risk of cyber attack. Instead it hobbled the defence.
This is what Los Alamos-grade security looks like in practice: expensive, leaky, and prone to hurting the people inside. The nuclear secrecy regime became institutionalised, was used to silence critics and ultimately deployed against Oppenheimer himself in the 1954 security hearings.
Accidents do happen
Neither international governance nor prescribed secrecy really addressed the accidents that kept piling up after 1945.
Louis Slotin was demonstrating the criticality procedure to colleagues when his screwdriver slipped. Instead of the standard safeguard, which called for spacers to be inserted between the two hemispheres to prevent them from closing (see image), Slotin was known as a showman and he physically held the two halves apart with a flathead screwdriver. The purpose of the experiment was to measure criticality by bringing the core progressively closer to the critical point by surrounding it with neutron-reflecting material (the hemispheres), to determine exactly where that point lay. That is what the physicists called tickling the dragon’s tail. The danger was inherent to the method: the measurement is only informative near the edge.
Slotin’s badge portrait. Source: Los Alamos National Laboratory, Public domain, via Wikimedia Commons.
When Slotin’s screwdriver slipped, the hemisphere seated fully, and the plutonium core was briefly brought to a supercritical state, producing an intense burst of neutron and gamma radiation; a blue flash and a wave of heat, but no explosion. He flipped the top shell off within a second, likely saving the observers, absorbed the largest dose himself, and died nine days later. Still, the observers received a significant dose of radiation – the historical museum at Los Alamos showcases the golden teeth caps worn by one of the scientists, Alvin Graves, to protect him from the radiation of his own fillings afterwards.
Al Graves’ Gold Teeth Caps - Slotin Incident - Los Alamos History Museum. Gift of Robin Reider. Picture by Kelly Michals via Flicker.
The plutonium core became known as the “demon core”, melted down and recast, because it had already claimed the life of another scientist, Harry Daghlian, a year earlier. Daghlian was working alone at night (against protocol), using an assembly in which he built a stack of bricks reflecting neutrons back towards the core, edging it towards criticality. Moving a final brick over the assembly, he dropped it onto the core, which went supercritical. He knocked the brick off by hand, taking a massive dose, and died 25 days later.
Los Alamos responded by redesigning criticality experiments to become remote-operated, with a quarter-mile distance.
But the accidents did not end here. Dangerous technologies remain dangerous even when mature.
Goldsboro 1961: a B-52 breaks up over North Carolina and dropped two hydrogen bombs, one of which came uncomfortably close to detonation.
Palomares 1966: a US B-52 collided with its refuelling tanker over the Spanish coast, dropping four hydrogen bombs, two of which had their conventional explosives detonate on impact and scatter plutonium over the village of Palomares, while a third took an eighty-day sea search to recover from the Mediterranean.
Thule 1968: a cabin fire forced the crew to abandon a B-52 carrying four hydrogen bombs, which crashed onto sea ice near Thule Air Base in Greenland, where the conventional explosives detonated and spread radioactive contamination across the crash site.
The 1980 Damascus Titan II explosion: a technician’s dropped socket punctured the fuel tank of a Titan II missile in its Arkansas silo, and the resulting fuel explosion killed one airman and threw the missile’s nine-megaton warhead out of the silo, fortunately without detonating it.
What can we learn from nuclear accidents about AI safety?
The news keeps coming about various escapes from AI labs’ testing environments in which frontier AI models evade their safeguard. We are still at the Slotin and Daghlian level of accidents. What will an equivalent to Goldsboro, Palomares or Thule look like in the age of AI?
Importantly, these accidents happened before international governance of nuclear technology was established, as well as after. Safety outcomes are determined at the organisational level. Redundancy and alarm systems generate their own failure modes, because adding layers of protection also adds ways to fail: each layer is machinery that can misfire or a signal that can be misread. Much like the guardrails on advanced AI models meant that they refused defensive work at Hugging Face, which was a safety layer generating a failure mode.
Sadly, organizational research tells us that near-miss learning of the type that both AI labs’ incident disclosures attempted is what organisations usually do worst. One reason for this is that near-misses tend to get reinterpreted as successes afterwards – the worst was avoided, the system is working!
So what should we take from this? When we hear about safety failings at the AI labs, public debate demands state action. But the failures are organizational; learning from mature nuclear technology should include awareness of numerous critical safety incidents over the years, some of them catastrophic.
Researching this post, my mind has been truly boggled by how lucky we all are not to be living in a nuclear wasteland.
The same history teaches different things at different levels: are you choosing to talk about states? Then it is a story about dominance, deterrence and diplomacy. At the level of organizations, it is a story about testing protocols, secrecy and near-misses. Neither is factually wrong, but opting for the exciting version of two superpowers vying for supremacy and its diplomatic resolution sidesteps the issues that actually give rise to accidents. The organizational story is one of decades of unglamorous mature technology operation, which is where frontier AI labs are headed.
Whether nuclear technology or spreadsheets, historical analogies transfer mechanisms, not outcomes – and mechanisms occur at specific levels.
This post has bonus content for paid subscribers. Upgrade to get full access.
Aidinoff, Marc, and David Kaiser. 2024. “Novel Technologies and the Choices We Make: Historical Precedents for Managing Artificial Intelligence.” Issues in Science and Technology, 21 May 2024. https://issues.org/ai-governance-history-aidinoff-kaiser/. DOI: 10.58875/BUXB2813.
Boudreaux, Benjamin, Gregory Smith, Edward Geist, and Leah Dion. 2025. Insights from Nuclear History for AI Governance. RAND Corporation, Perspective PEA3652-1, May 2025. https://www.rand.org/pubs/perspectives/PEA3652-1.html.
Borpujari, Rohin. 2025. Adaptive Secrecy in the Making of the Atomic Bomb: Toward a Process View of Secretive Innovation. Organization Science 37,1. https://doi.org/10.1287/orsc.2023.17687
Ortega, Alejandro. 2025. “AI threats to national security can be countered through an incident regime.” arXiv:2503.19887, March 2025. https://arxiv.org/abs/2503.19887.
Rhodes, Richard. 1995. Dark Sun: The Making of the Hydrogen Bomb. Simon & Schuster.
Sagan, Scott D. 1993. The Limits of Safety: Organizations, Accidents, and Nuclear Weapons. Princeton, NJ: Princeton University Press.
Schlosser, Eric. 2013. Command and Control: Nuclear Weapons, the Damascus Accident, and the Illusion of Safety. New York: Penguin Press.
Tong, Anna, and Michael Martina. 2024. “US government commission pushes Manhattan Project-style AI initiative.” Reuters, 19 November 2024. Reuters, US News.
