Who Gets to Turn Down the Sun?
Elon Musk has now proposed something that sounds less like climate policy than the opening scene of a science-fiction film. Humanity, he suggests, could ultimately regulate the temperature of the Earth using "sentient satellites" positioned around the Earth-Sun Lagrange points, with the machinery launched from the Moon by electromagnetic mass drivers. Musk claims that such a system could solve global warming for roughly a billion years. He has also argued that ordinary decarbonisation will eventually be insufficient against the much larger natural threats confronting life on Earth.
There is an important qualification before anyone reaches for the apocalypse button. Musk has not announced that SpaceX is about to construct a planetary thermostat, and the remark is better understood as a speculative technological vision than a funded engineering programme. Yet the underlying physics is not something Musk invented after an unusually imaginative breakfast. Scientists have been discussing space-based solar geoengineering for decades, including vast clouds of small spacecraft or sunshades positioned near the Sun-Earth L1 region to reduce slightly the amount of solar radiation reaching the planet.
The principle is almost embarrassingly simple. Earth receives energy from the Sun and radiates energy back into space. Change that balance sufficiently and the planet's average temperature changes. Greenhouse gases warm the lower atmosphere by altering the outgoing side of the balance. A sufficiently large space-based sunshade would attack the other side by reducing incoming sunlight.
The required reduction need not mean putting Earth into anything resembling darkness. Published studies of L1 solar geoengineering have examined reductions of around 1.7 to 1.8 per cent of incoming solar flux as sufficient, in particular modelled circumstances, to counteract warming associated with a doubling of atmospheric carbon dioxide. One proposal envisaged not one colossal science-fiction umbrella but a cloud of vast numbers of tiny autonomous spacecraft near L1.
That is where Musk's "sentient satellites" presumably enter the picture. Instead of building a single rigid shield hundreds or thousands of kilometres across, humanity could eventually deploy an enormous distributed constellation whose elements altered their orientation or position to regulate the amount of solar energy reaching Earth. Artificial intelligence could coordinate the swarm, monitor climatic conditions and continually adjust the system.
The Earth-Sun L1 point is attractive because it lies between Earth and the Sun, roughly 1.5 million kilometres from Earth. A spacecraft operating around that region can remain approximately aligned with the Earth and Sun while requiring comparatively modest station-keeping. It is not, however, a celestial parking lot into which millions of satellites can simply be dumped and forgotten. L1 is dynamically unstable. Spacecraft operating there require continuing corrections.
Musk adds another futuristic element: manufacture the satellites or their components on the Moon and throw them into space using mass drivers. A mass driver is essentially an electromagnetic launcher. The Moon is an attractive location because it has only about one-sixth Earth's surface gravity and virtually no atmosphere. Once an industrial civilisation had established mining, manufacturing and large-scale electrical generation on the lunar surface, electromagnetic launchers could conceivably propel enormous quantities of material into space without burning chemical rocket fuel for every kilogram.
We should therefore distinguish the extraordinary scale of Musk's proposal from the underlying physics. Space sunshades are not perpetual-motion machines. They do not violate any obvious law of nature. Peer-reviewed studies have investigated them, and researchers continue to publish proposed development roadmaps. The extraordinary part is constructing, transporting, controlling and governing something large enough to modify the energy balance of an entire planet. And that final word, governing, is where the really interesting problem begins.
Suppose Musk is right. Suppose that by 2075 humanity possesses lunar mines, mass drivers, cheap space transportation, extraordinarily capable artificial intelligence and millions of autonomous spacecraft capable of regulating incoming solar radiation. We would have acquired something unprecedented in human history: an adjustable planetary thermostat. Who gets the thermostat? That question is considerably more difficult than the engineering problem.
If the satellites reduced incoming sunlight too much, Earth would cool. Reduce it too little and the intended warming offset disappears. Change the distribution of solar radiation and effects would not necessarily be uniform across regions and seasons. Climate is an immensely complicated coupled system involving oceans, atmosphere, ice, vegetation and circulation patterns. Controlling global mean temperature is not the same thing as controlling every consequence of the intervention.
The problem is made still more interesting because reducing sunlight does not reverse everything caused by a supposed increased atmospheric carbon dioxide. A sunshade could counteract some warming while leaving elevated atmospheric CO2 in place, including effects such as ocean acidification. Solar geoengineering is therefore not equivalent to removing greenhouse gases from the atmosphere. It would be an intervention into one component of the planetary energy system to compensate for changes in another.
Now add artificial intelligence. The popular fear is obvious: the AI hallucinates, makes a calculation error and accidentally turns Earth into an ice ball. That formulation is probably too cinematic. A properly engineered planetary system would never rationally give one language model an unrestricted instruction reading, "Please maintain a pleasant climate." Safety-critical engineering uses independent sensors, redundant systems, bounded control authority, verification, human supervision and multiple layers of fail-safe mechanisms.
A well-designed sunshade swarm should also be physically incapable of changing Earth's climate instantaneously. The system could be divided into millions of independently controllable elements, with strict limits on how rapidly total shading could change. If something went wrong, the individual shades could rotate edge-on to the Sun, disperse or otherwise reduce their effect. Recent planetary-sunshade proposals explicitly emphasise reversibility.
But eliminating the cartoon version of the AI risk leaves a much more serious problem. Complex automated systems can fail without hallucinating. Sensors can produce erroneous readings. Software can contain bugs. Communications can fail. Models can be wrong. Components can degrade. Solar storms can damage electronics. Different control systems can interact unexpectedly. An adversary can attack software deliberately. Anyone who has watched an operating system install an unwanted update should hesitate before connecting software to the Sun.
The timescale makes Musk's billion-year claim especially revealing. No human technological system has operated continuously for anything remotely approaching that period. Homo sapiens has existed for only a few hundred thousand years. Agriculture is roughly twelve thousand years old. Industrial civilisation is a few centuries old. Modern digital computing has existed for less than a century. Yet we are contemplating machinery supposedly regulating planetary climate on geological timescales.
Who repairs it in the year 12,026? Who repairs it in 102,026? What political institution survives long enough even to remember why it was constructed? Technology does not abolish maintenance. It institutionalises maintenance.
There is then the cybersecurity problem. A system capable of changing Earth's temperature would instantly become perhaps the most strategically important infrastructure humanity had ever constructed. Hacking a bank steals money. Hacking an electricity network can black out a city. Hacking a planetary solar-control network could potentially alter environmental conditions for continents.
Such a system could never sensibly be controlled by one company, one country or one artificial intelligence. Yet distributing control creates its own difficulties. Does America receive one vote and China another? Does India, with more than a billion people, receive the same authority as Luxembourg? Do countries contribute according to population, GDP or exposure to climatic effects? What happens when Russia wants the planet 0.3 degrees warmer and India wants it 0.3 degrees cooler?
This is not fanciful. Countries do not experience climate identically. A temperature regarded as damaging somewhere might produce agricultural advantages elsewhere. Changes in rainfall could help one region while harming another. Once humanity can deliberately alter the planetary climate, weather ceases to be entirely an act of nature and begins acquiring political responsibility.
Imagine a catastrophic drought occurring after the satellite system has been adjusted. Even if the adjustment did not cause the drought, millions of people would believe somebody had done it to them.
The geopolitical consequences would be extraordinary. A country suffering failed monsoons might accuse the countries controlling the sunshade of climate warfare. Another experiencing floods might demand that the system be adjusted. Scientists would produce competing models. Governments would accuse one another of manipulating the modelling assumptions. Every hurricane, heatwave, drought and failed harvest would acquire a new suspect: the planetary thermostat. The problem is therefore not simply that artificial intelligence might make a mistake. Humans might disagree about what counts as a mistake.
Musk's terminology inadvertently reveals the danger. "Sentient satellites" suggests autonomous intelligence managing a planetary system because the complexity exceeds ordinary human administration. That is precisely when the political question becomes most acute. The more complicated the system becomes, the easier it becomes for its operators to tell everyone else that the experts and algorithms understand it and ordinary people do not.
We encountered the same issue with data centres on a vastly smaller scale. Technological necessity becomes a political argument. AI requires electricity, therefore data centres must be built; national competitiveness requires AI, therefore communities must accommodate the data centres. With planetary geoengineering the logic could become almost absolute. The climate model requires a particular reduction in solar radiation; therefore the satellites must implement it. Political disagreement becomes interference with the scientifically determined optimum. At that point technocracy has escaped the planet.
There is also a moral-hazard problem. It assumes totally uncritically the climate change alarmist agenda. That is something to be proved not assumed, with bodies such as the UN IPCC moving away from its past climate change alarmism. Musk seems to still be in that past. Another article at the blog today goes into this.
https://redacted.inc/2026/09/musks-radical-plan-to-cool-earth/
