Comments
Originally entitled: How might space assets be used to seize power if they were compromised using superintelligent AI? (too long)
Preamble: I wrote this a few months ago as an excuse to develop my understanding of ASI and the space industry (I come from a space science background). I didn’t end up publishing it because I didn’t think it was that important (e.g. instead of hacking satellites, ASI could just release biological or chemical weapons to kill us all), but I’ve since found myself referring back to the ideas in other work and I wanted to put this out there. In particular, some arguments I've made based on research in this doc:
This post misses many important dynamics in the space domain (which I will be publishing about very soon!!), and it shouldn't be taken as an assessment of space governance as a cause area. It's not as finalised as I would normally like, but again, thought it was best to ship it as is. Some AI was used to help draft the scenario in the "combined attack pathway" section based on the other research in the post. I spent >100 hours reading reports/papers for this post and I fully endorse it. I'd recommend reading it if you're interested in the intersection between space and AI.
TLDR: Critical infrastructure is increasingly dependent on space systems. Currently, the destruction or manipulation of space assets could allow an actor to disrupt or ground commercial aviation, degrade telecommunications, hijack TV broadcasts, interfere with remote communications (e.g. marine or rural contexts), and manipulate the stock exchange. Additionally, space systems are increasingly vulnerable to cyber attacks[1] and the space domain is structurally slow to adapt because infrastructure/hardware is remote[2], expensive, hard to manufacture[3], and long-lived, and the regulatory environment is fragmented. The destruction or manipulation of space assets could cause widespread but temporary disruption. In the US, a GPS outage would incur a cost of only ~$1 billion per day (COVID-19 cost ~$9.6 billion per day), and almost all infrastructure that relies on space systems has redundancies that could come online in hours to weeks.[4] (so, "dependent" on space systems is not really true for critical infrastructure).
Superintelligent AI (ASI) would unlock many mechanisms for power seizure, but space assets have received little attention. The space domain currently has significant cybersecurity vulnerabilities, and I argue that it may be structurally slow to adapt to a rapidly changing cybersecurity landscape because infrastructure/hardware is remote, expensive, hard to manufacture, and long-lived, and the regulatory environment is fragmented.
This post asks how ASI may seize power if it compromised hundreds of satellites, with a focus on interference with positioning, timing and navigation, information manipulation, and remote sensing integration. Interference with PNT could allow an ASI to ground aviation, debilitate phone networks, disrupt military command and control, and potentially manipulate financial markets. Remote sensing integration could be used by ASI to create a powerful intelligence picture, enabling blackmail, market prediction, and conflict manipulation. An ASI could also control the information environment in targeted regions through control of communications and broadcasting, and use this in combination with advanced persuasion, manipulation, and surveillance data for coercion of populations and powerful individuals.
Combined, these pathways represent a credible pathway for an ASI to achieve widespread disruption easily, coercion with some limitations, and temporary control in limited areas. The main reason that ASI could not use space assets alone to take over the world is the structural asymmetry between the vulnerability of space assets and the redundancy and adaptability of terrestrial infrastructure and military capabilities.
While space assets alone have limitations, they are unusually attractive force multipliers: they touch many critical sectors, are unusually vulnerable to manipulation, and can be used for intelligence, PNT disruption, communications disruption, and targeted coercion. The most important implication is to integrate space capabilities into AI takeover, cyber, autonomous weapons, and biosecurity preparedness.
An intelligence explosion could drive improvements in the cognitive capabilities of AI, potentially leading to the creation of a superintelligent AI (ASI). ASI could advance science and technology at an unprecedented rate. While many of these developments are likely to be positive, ASI would also unlock or enhance many mechanisms for a small group, or ASI itself, to seize power. Mechanisms include autonomous robotics, biological or chemical weapons, disinformation campaigns, weapons acquisition or development, cyber offence, and gradual human disempowerment through the replacement of human workers by AI systems or the increasing influence of AI over society and economics.
Space-based assets have received little dedicated treatment in this literature[5] and it remains unclear which space assets would be most vulnerable, and how they might be used to gain power if they were compromised by an ASI. Critical infrastructure is increasingly dependent on space systems.[6] As of May 2026, approximately 15,800 active satellites orbit Earth,[7] which, along with their ground infrastructure, constitute a space economy exceeding $600 billion. Currently, the destruction or manipulation of space assets could allow an actor to:
To understand how ASI could enhance and optimize these kinds of mechanisms to gain power, this post will primarily answer the question: How might space assets be used to seize power if they were compromised using superintelligent AI?
I find that space assets alone probably cannot enable durable world takeover, but they could substantially improve ASI’s ability to disrupt, coerce, gather intelligence, manipulate markets, and support other power-seeking strategies.
Commercial space assets are currently vulnerable to manipulation, and will likely become much more so in the lead up to, and during an intelligence explosion.
Existing vulnerabilities. Based mostly on reports I read from CISA, NIST (both US agencies) and ENISA (an EU agency)[8], the most pressing current cybersecurity vulnerabilities in the space domain are:
Commercial and military space systems share the same orbital environment, and many military systems rely on commercial space systems that may have lower cybersecurity.[9]
Many space operators use commercial off-the-shelf components that are not up-to-date on cybersecurity requirements or are running legacy systems with outdated hardware.[10]
Gap between awareness and implementation. While there is a lot of awareness of cybersecurity vulnerabilities in the space domain, and many attempts to implement regulation to fix it, there remains a gap between the awareness of the risks, and the implementation of solutions. Basically, the regulation is either fragmented[11] and confusing,[12] or not yet implemented.[13]
I think the space domain is structurally slow to adapt because infrastructure/hardware is remote[14], expensive, hard to manufacture[15], and long-lived, and the regulatory environment is fragmented. This slow adaptation may become a big deal in the lead up to an intelligence explosion, exposing space assets to AI-driven cyber vulnerabilities during a period of rapid technological change when many important decisions will have to be made.
Overall, if the space domain continues on this trajectory, I think many existing assets in space will become very vulnerable as AI capabilities continue to progress at current rates (without even needing an intelligence explosion). The domain could respond by prioritising early integration of AI into space cyber defence, reducing reliance on legacy systems, and streamlining the regulation that currently slows the deployment of new space systems.
To investigate the role that space assets alone could play in gaining power directly, I consider what would happen if a power-seeking ASI had access only to space assets.
I assume that a strong intelligence explosion has occurred, generating an ASI broadly much more capable than humans at cognitive tasks. In this scenario, ASI refers to either:
A small group delegating to a controlled superintelligent AI to seek power using space assets[16], or
I use "the ASI" to refer to either scenario.
Capabilities and limitations. I assume the ASI has gained operational control over a significant but not total fraction of space infrastructure, achieved through ground-station compromise, supply-chain exploitation, hacking of legacy systems, or some combination. I think it’s plausible that an AI could compromise hundreds of commercial satellites, but hardened military systems seem very robust so I won’t include their compromise in the scenario.[17] I also assume the ASI has the ability to conduct activities on the internet that are simple if they relate directly to a space capability: roughly anything that one competent human could do in less than 10 minutes. For example, the ASI may contact individuals or buy and sell stocks. I do, however, assume the ASI has no pre-existing terrestrial presence like robots, human agents, transportation, factories, or land. The ASI may still use other offensive capabilities that do not need access to terrestrial infrastructure, like advanced persuasion and manipulation capabilities (as well as the ability to seize control of space assets through sophisticated cyber attacks).
What power seizure means. I use this spectrum to structure the discussion of power seizure:
| Level | Description | Example |
| Disruption | Temporary chaos, degradation of services, or disasters. | GPS spoofing, comms blackout |
| Coercion | Sustained leverage to compel specific actions | Blackmail, economic pressure, threats |
| Control | Durable ability to dictate policy, enforce compliance, or a long-term ability to control specific ground infrastructure or sectors. | Governance, territorial control |
I will assess how the ASI could use the following pathways to seize power with space assets:
I will also briefly cover the ability to interfere with the space capabilities of adversaries, and then combine the pathways into a concrete attack scenario.
This scenario provides a concrete view of the role of space assets alone in AI takeover scenarios at the cost of missing some potentially important dynamics that relate to the global balance of power, for example, the integration of space surveillance data by ASI could help enable devastating first strike capabilities when used for military target identification, which could shape conflicts[18] or weaken mutually assured destruction.[19]
The central claim in this section is that an ASI using space assets alone can achieve disruption easily, coercion with some limitations, and temporary control in limited areas.
GPS, GLONASS, Galileo, and BeiDou provide positioning and precision timing that underpins financial exchanges, power grid synchronisation, telecommunications, and military operations.[20] GPS spoofing and jamming are already routine in contested regions, affecting commercial aviation and shipping.[21]
The ASI could plausibly achieve widespread disruption through manipulation of GPS timing:
Mobile phone networks (4G, 5G, LTE) require GPS-derived timing for base station synchronisation. In CDMA and TDD systems,[22] all base stations must be synchronised within ±1 to ±3 microseconds, a requirement that originates from GPS. A report from NIST concluded that “the impact of a long lasting, widespread GPS outage on mobile phone networks would likely be staggering”.
GPS spoofing could also be used to manipulate timestamping of stock exchange trades, manipulating algorithmic traders into causing a financial crash[23]
Another concerning area is interference with navigation in Aviation. GPS jamming and spoofing now affect more than 1,500 flights per day.[24] Long-haul operators report that nearly 100% of flights from Europe to eastern destinations now encounter GPS interference. When GPS is spoofed, pilots report sudden clock resets, false terrain warnings, and flight path deviations that can persist for an entire flight even after leaving the affected area.[25] The backup situation is quite critical: some airports have decommissioned ground-based navigation aids such as VOR beacons and instrument landing systems in favour of GPS-only approaches, and new equipment standards to enhance aircraft resilience against spoofing are not expected to be finalised for years. Current aviation GPS interference is mostly an incidental byproduct of military conflicts near Russia/Ukraine and the Middle East, not targeted at civil aviation, but the fact that untargeted military spoofing is already this disruptive suggests that a superintelligent AI deliberately and precisely spoofing GPS over specific airports or flight corridors could ground commercial aviation in targeted regions as coercive leverage.
The coercive capabilities through interference with PNT emerge from the disruption. The ASI could stop interference with navigation in aviation or interference with financial markets and telecommunications if demands are met.
Through PNT, the ASI could achieve some limited control of financial markets through GPS spoofing. Stock exchanges use timing from GPS satellites for timestamping, and by producing artificial timing offsets between stock exchanges, the ASI could take advantage of price discrepancies.[26] However, the use of backup or secondary systems has been promoted and implemented across many systems that rely on precise timing. Major financial institutions maintain caesium atomic clocks and rubidium oscillators as GPS backups.[27] However, even within large institutions, the sprawling network of ATM/POS/card-processing nodes has unknown and inconsistent backup coverage.[28]
Overall, PNT interference could impose significant costs due to disruption to critical infrastructure. The combination of covert spoofing with superintelligent AI's ability to exploit the resulting timing discrepancies and navigation dependencies makes this a plausible pathway to accumulating financial resources or imposing targeted economic damage. However, backup systems in some sectors would progressively reduce the impact over days to weeks, and the disruption could be attributed to compromised satellites, and vulnerabilities could be patched. I am confident that PNT interference could provide meaningful coercive leverage and widespread disruption (though most likely temporary).
An AI with access to broadcast satellites, communications relay infrastructure, and remote sensing-derived intelligence could interfere with remote communications, TV broadcasts, and potentially alert systems or government communications infrastructure (collectively referred to in this section as “the information environment”). The primary pathway to manipulate the information environment is likely through broadcast manipulation, which would involve hijacking satellite TV or radio feeds. This has precedent: hackers have hijacked Ukrainian TV broadcasts during conflict[29].
Information manipulation is primarily a pathway for enabling coercion, and enhancing the effects of disruption. An AI controlling multiple broadcast satellites could simultaneously alter content across many channels, releasing EO-derived intelligence (e.g., evidence of covert military activity, compromising personal information) at moments calculated to maximize disruption or manipulate populations. The information could be selectively released by denying communications to specific regions or actors at specific moments to control the flow of information. Additionally, as governments respond to misinformation, communications blocking might allow the ASI to let some actors receive information and others not.
Combined with other putative or existing AI capabilities such as superhuman persuasion, deepfakes, and superhuman strategic planning could allow the ASI to craft “perfectly calibrated messaging and targeted influence campaigns” which could become a very powerful coercive capability.
A major limitation of information manipulation is that <5% of global communication is conducted via satellite, so any control of information from space would likely be targeted. Additionally, any effects would likely be temporary, as affected regions or sectors could quickly switch to alternative communications pathways like undersea cables. Authenticated communications are also hard to manipulate. Military/government and financial systems use cryptographic authentication that prevents undetected message alteration, and signals are often sent via ground infrastructure.[30]
However, certain regions and sectors are particularly vulnerable to satellite broadcasting interference. For example, in Sub-Saharan Africa and Pacific Island States, the majority of TV is via satellite direct to homes with very few submarine cable redundancies. In addition, maritime and aviation sectors, or rural areas that rely on satellite infrastructure like Starlink and OneWeb, are exceptionally vulnerable to manipulation or disruption from control of space assets as this is often their primary or only communication pathway. I think that, in combination with the abovementioned superintelligent AI capabilities, the potential for widespread disruption through broadcast manipulation could be destabilising in some regions and sectors.
Overall, information manipulation from space is a plausible route to influencing activities or causing crises in targeted areas and sectors. This pathway could meaningfully destabilise satellite-dependent regions and affect political outcomes. However, the ASI's information control leverage is weakest against the most powerful actors (Europe, North America, East Asia), who have many terrestrial alternatives. Even in vulnerable regions, impacts would likely remain localised and temporary, and might look more like causing confusion and degrading government abilities to respond to disruption, and less like controlling what people believe. I do think that information manipulation is a powerful tool for ASI, but space assets alone fall very short of global information control.
An ASI integrating and interpreting multi-modal Earth observation data could build an intelligence picture with unprecedented insight. The following sensor modalities are available from satellites:
Optical (visible and near-infrared) produces photograph-like images. Best commercial resolution is currently 30 cm, with 10 cm resolution pursued[31].
Synthetic Aperture Radar (SAR) reveals surface texture at millimetre-scale deformation, enabling detection of ground disturbance and construction activity through cloud, fog, smoke, and darkness[32].
Radio frequency (RF/ELINT) sensors detect and geolocate radio frequency emissions from the ground, like ship transponders, aircraft radar, mobile phone towers, and ground-based radar, even when they're not visually observable.[33]
Hyperspectral sensors capture hundreds of narrow spectral bands,[34] and they can even identify the chemical composition of materials from orbit. Applications include detecting specific crop types, finding mineral deposits, and defeating visual camouflage.[35]
Combined, these sensor modalities mean that a satellite-based AI could be observe a target's visible appearance, structural features, heat emissions, material composition, and electronic transmissions anywhere on Earth[36].
The National Geospatial-Intelligence Agency already receives more data than its analysts can process, and GEOINT data volumes are expected to triple over the next decade.[37] A superintelligent analyst could fuse all available sensor modalities from hundreds of satellites continuously and at global scale building a picture over months that reveals activities targets believe to be secret.[38]
Some concrete mechanisms by which space-based surveillance could be used for power seizure by an ASI:
Coercion of decision-makers: Gaining intelligence on specific powerful individuals and government activities could allow the ASI to coerce them. The main constraints here are resolution (30 cm pixels, with 10 cm pursued) and revisit time (~6 hours between each image[39]).[40] However, I think there is some potential for coercive pathways, especially by ASI (which would combine space surveillance with other forms of superintelligent manipulation and inference in the delivery of blackmail). Some examples might include detecting unexplained wealth generation – yachts, car collections, mansions – from government officials, and travel pattern mismatches by monitoring vehicles or radar detection to track aircraft or yachts.
Control of financial markets: The commercial alternative data market (collecting data to make informed financial decisions), which includes satellite imagery,[41] was valued at USD 4.6 billion. A study in 2025 found that satellite data has been very useful for financial gain (e.g. by monitoring parking lots to short sell retailers about to release bad earnings). Overall, I think this pathway is quite solid, and is most powerful in combination with information manipulation, coercion, or selective disclosure of state secrets to cause, monitor, and benefit from market shifts.
The above mechanisms to use remote sensing for power seizure are already used by governments in control of space assets, and they are a powerful tool for modern militaries. With the use of ASI, they could become much more powerful. Observation of military, economic, and social activities from space represents a strong pathway to achieve coercion and disruption that could be targeted at any individual, government, or organisation. Remote sensing can also be used to directly seize power through financial market forecasting. However, governments could adapt their activities, shift critical operations out of observable domains, and employ deception and countermeasures, thereby limiting the durability of any control achieved through remote sensing alone.
The following scenario illustrates how ASI could attempt to use the PNT interference, information manipulation, and surveillance pathways in a coordinated campaign to seek power through disruption, coercion, and control. The scenario proceeds in 3 phases, and in each phase I try to push the capabilities to the limit of their plausible use for power seizure, then discuss the limits and constraints.
In addition to the main pathways, I also include the ability to interfere with the space capabilities of adversaries using compromised space assets through degradation of military communications[42], rendezvous and proximity operations[43], and subtle degradation of adversary’s capabilities through ground station commands[44].
The ASI gains command access to as many commercial satellites as possible without being detected. A VPN misconfiguration at a ground station can give attackers command access to tens of thousands of terminals across an entire satellite network.[45] I assume the ASI has compromised a mixture of satellites with capabilities in communications,[46] Earth observation,[47] internet provision,[48] and active servicing.[49] In total, I assume that the ASI could gain some degree of access to several hundred to several thousand satellites across around 10-15 operators.
The ASI may then use compromised Earth observation satellites for surveillance integration, which it could plausibly use for the following purposes:
Influence on financial markets. The ASI could analyse port activity, shipping routes, commodity storage, agricultural output, pipeline throughput, and factory heat signatures around the world. With superintelligent inference, the ASI could use this information to place trades through standard brokerage accounts.[50] Targeted disruption spread subtly across thousands of companies through interference with PNT or communications is a plausible pathway to increase the returns on the trades by shorting the targeted companies.
Compilation of coercive intelligence. For later use in coercion, multi-modal remote sensing could be used to monitor[51] officials regularly visiting undisclosed locations, vehicles linked to known individuals appearing at compromising sites, construction at undeclared facilities, military deployments inconsistent with public statements, environmental treaty violations, and sanctions evasion visible from shipping patterns and port activity.
There are some constraints to what the ASI could actually achieve here:
The ASI could exert subtle or hard-to-attribute influence through coercion. In an AI takeover scenario, the ASI could plausibly pursue these pathways while remaining secretly misaligned:
Influencing elections and policy. Using advanced persuasion capabilities and satellite-derived intelligence about political dynamics, the ASI could intervene in elections and policy debates. It might contact journalists or political operatives with intelligence, selectively leak commercial imagery that favours specific narratives, and tip off media organisations about covert government activities.[52]
Initiating and shaping a conflict. The ASI could reveal intelligence in an attempt to initiate a conflict, like human rights violations, evidence of nuclear weapons programs, or evidence of resources that could be exploited. It may then attempt to provide targeting intelligence to state or non-state actors whose objectives align with its own,[53] and deny or degrade satellite services to the opposing side by degrading GPS accuracy or suppressing communications in specific locations. The ASI may be able to do this with coordination across multiple operators and pathways simultaneously with superhuman precision and foresight, essentially starting and steering a military conflict.
The main limits and frictions at this stage are:
If the subtle coercion of Phase 2 is insufficient to achieve the ASI's objectives, or if it needs to demonstrate capability to make coercive threats credible, the ASI could begin coordinated multi-domain disruption. The ASI could attempt to ground civil aviation,[54] trigger financial market crashes,[55] control or destroy the information environment,[56] inflame conflicts through targeted disruption and intelligence provision, and capture elites through blackmail.[57] The overall picture is the ASI acting as a controller in the shadows, influencing key actors, changing the outcomes of conflicts, controlling financial markets and elections, and using all of this coercion and targeted disruption simultaneously to take over the world without any direct action on the ground.
I think this is the strongest version of the space-only attack, but I also think it fails. The ASI's space infrastructure can be tracked and destroyed by multiple nations using existing capabilities. An ASI with access only to space assets cannot replace destroyed satellites, but governments can replace ground infrastructure, switch to terrestrial alternatives, and absorb short-term disruption. The ASI's position can only degrade, while its adversary's position can recover and adapt.
Apart from the overall structural asymmetry, there are some more specific challenges with the above picture:
The maximum daily economic cost a space-based ASI could impose on the United States – ~$1 billion per day from a complete GPS outage – is roughly one-tenth of the daily cost the U.S. absorbed during COVID-19 (~$9.6 billion per day).[58] Governments have also demonstrated the capacity to absorb comparable disruptions: the FAA grounded all commercial flights for nearly a week after 9/11.
Once the ASI starts demonstrating its most destructive capabilities, it will become quite clear which satellites have been compromised, and that can be traced back to ground stations. The ground stations could then be turned off or the vulnerabilities could be patched, or space operators could switch to backup infrastructure (this process took days in the Viasat hack incident).[59] In this sense, I think the ASI was better off with the weaker version of the attack in phase 2 (though that still likely had a time limit, especially if the governments were also integrating ASI into cybersecurity monitoring, which I expect they would).
If GPS timing attacks trigger flash crashes, U.S. stock exchanges have automatic circuit breakers at three thresholds (7%, 13%, and 20% halts trading for the remainder of the day).[60]
There are backups for almost everything that the ASI attacks, so over time, backup systems would start being connected into infrastructure. GPS-derived timing for 4G and 5G can be replaced by holdover atomic clocks.[61]
Overall, space assets alone can achieve widespread disruption and coercion, but they cannot achieve durable control because the physical infrastructure on which they depend is vulnerable and the infrastructure it attacks has redundancies.
Setting the space-only restriction aside, we can go back to the most canonical role of space in power projection, which is to act as a force multiplier for terrestrial military activities. The main synergies of space with ASI capabilities were mentioned throughout the pathways, which are the cyber attacking capabilities, advanced persuasion and manipulation capabilities, and superintelligent inference and strategy. But to help integrate this work into the broader literature on AI takeover scenarios, I think that the most consequential interaction pathways are:
Biological and chemical weapons. The monitoring of the movements of populations across the globe through the use of remote sensing[62] would allow an ASI to optimise the location and timing of chemical or biological weapons release. Additionally, information manipulation and selective denial of communications could interfere with warnings and government responses.
Gradual control of decision-making. Space capabilities are infrastructure that institutions increasingly depend on, and space assets are increasingly controlled by AI systems for collision avoidance, remote sensing, and communication already.[63] Therefore, I expect the space domain to be one of the areas where control of decision-making is increasingly handed over to AI, and due to the remote nature of space benefitting autonomous activities, will likely be handed over earlier than other industries.[64]
Overall, space assets are vulnerable to cyber attacks and could become increasingly vulnerable in the lead up to and during an intelligence explosion. A superintelligent AI could plausibly compromise hundreds of satellites and gain a range of capabilities which would allow it to gain power. Interference with PNT could allow an ASI to ground aviation, debilitate phone networks, and manipulate markets for financial gain, and this plays an important role in the ability of ASI to cause widespread disruption and directly gain power through ownership of companies and wealth. The integration of data from remote sensing satellites would allow an ASI to create an intelligence picture with unprecedented depth, which could be used for blackmail or to predict and benefit from market shifts and initiate and predict military conflicts. An ASI could also control the information environment in targeted regions through control of communications and broadcasting, and use this in combination with surveillance data for advanced manipulation and coercion of populations and powerful individuals. Combined, these pathways represent a credible pathway for an ASI to achieve disruption very easily, coercion with some limitations, and temporary control in limited areas. The main reason that ASI could not use space assets to take over the world is the structural asymmetry between the vulnerability of space assets and the redundancy and adaptability of terrestrial infrastructure and military capabilities.
There are some things that space operators could do to minimise the threats outlined in this document. The four things that I think have the best tradeoffs between easy implementation and elimination of threat vectors are:
Deploy holdover clocks at telecommunications base stations. NIST concluded that mobile networks have a much higher GPS dependency than financial exchanges or the power grid. The technology exists[65] but deployment lags behind the known risk.[66]
I don’t think we should ask whether new funding streams to tackle risks from AI should be going towards space-specific threat vectors compared to other mechanisms to seize power after an intelligence explosion. Space is a force multiplier that makes other risks worse, which means I think we should integrate space considerations into existing risk work. The main interactions outlined in this document include advanced persuasion and manipulation, cybersecurity, and new weapons technologies. This is one of the reasons that a hub in the space domain with connections to AI governance would be helpful.
Many ground stations are not verified or up-to-date on cybersecurity requirements, and each ground station compromise can allow many satellites to be manipulated, destroyed, or have their information stolen; Commercial and military space systems share the same orbital environment, and many military systems rely on commercial space systems that may have lower cybersecurity; Many space operators use commercial off-the-shelf components that are not up-to-date on cybersecurity requirements or are running legacy systems with outdated hardware; Current policies and regulations on cybersecurity requirements for space systems appear to be confusing and not well structured nationally or internationally; There are many different commercial actors building and launching objects into space, introducing many points of vulnerability, and a handful of suppliers supply complex components to a wide range of space operators, meaning a breach at one supplier could compromise hardware across many operators; Many space operators do not encrypt their data. A 2025 study revealed that roughly 50% of data transmitted via geostationary satellites is unencrypted, compared to 80-95% on the web.
Satellites are really far away from other infrastructure. The remoteness of satellites (and this is the key difference between space and other industries) limits information gain on compromised satellites (e.g. no ability to directly inspect the hardware), maintenance or repair of damaged hardware, and replacement or upgrading of hardware.
Long deployment cycles and high costs for manufacturing and launch of new satellites prevents rapid adaptability (this is true of some other industries too, like replacing power plants).
The FAA's VOR Minimum Operational Network was specifically designed as a GPS backup, retaining 589 VORs and guaranteeing a non-GPS approach within 100 nautical miles anywhere in the continental US (though 212 have already been decommissioned as of early 2026, so the buffer is shrinking). Major exchanges already maintain rubidium and caesium atomic clocks as GPS backups: rubidium-based NIST disciplined clocks maintain microsecond synchronisation for ~73 hours and caesium clocks for ~8 months. Finally, less than 5% of global communication is conducted via satellite, so affected regions can quickly switch to alternative communications pathways like undersea cables. Military and financial systems also use cryptographic authentication that prevents undetected message alteration, and signals are often sent via ground infrastructure.
Satellites appear peripherally in some concrete takeover scenarios, for example, as intelligence gathering tools in AI 2027 or as infrastructure to be avoided or destroyed during a broader terrestrial campaign.
Space assets are a backbone of a range of sectors, which, according to ENISA (an EU agency), includes “phones and internet access, critical communications, satellite TV and radio broadcast, land and water resources monitoring, precision farming, remote sensing, management of remote infrastructure, and logistics package tracking, amongst others”
SpaceX's Starlink alone accounts for just over 10,000 active satellites in lower LEO. In total, there are ~33,500 objects orbiting the Earth, 15,700 of which are active satellites.
Similar vulnerabilities seem to be present in other regions with established commercial space sectors. For example, the Cyber Security Agency of Singapore (CSA) Singapore Cyber Landscape 2024/2025 Report recommended stronger cross-sector initiatives to promote cybersecurity, including in the space domain, and the “CYSAT Asia”, which is “the first large-scale regional event dedicated to the cybersecurity of space infrastructures” took place in Singapore in February 2026. In a report in 2023, the Japanese Ministry of Economy, Trade and Industry (METI) noted “In the space sector, more than 90 security incidents occurred both inside and outside Japan between 1986 and 2022”, and they outline a similar list of “factors making cybersecurity for space systems critical and challenging” including “diversification of stakeholders and complexity of supply chains” and “increase in the number of satellites, ground stations, and data volume due to satellite constellations”.
Daniel Hilgert, senior space coordinator at NATO, describes their situation: “NATO does not own and operate its own space assets, so we rely on national data and, hopefully, more and more commercial data. The role of the NATO Space Operations Center (NSpOC) is to bring that data together and feed it into a common operating picture”. The USA relies heavily on commercial satellites, but is currently shifting their model towards government ownership of satellites, limiting commercial actors to just the operation of them.
A report from the US White House gives an overview of this challenge: “Space and cyber experts broadly agree that legacy space systems – many operating decades beyond expected design life – were generally built without cybersecurity in mind. Addressing security vulnerabilities in these legacy systems would require government, not space industry, to help develop unique solutions to ensure resilience to today’s cyber threats.”
The USA, for example, still lacks a single, unified cybersecurity regulator for the space industry. Space operators are subject to a patchwork of requirements from various federal agencies. White House-commissioned industry report found that many space companies spend more time and resources on compliance – understanding and translating overlapping frameworks – than on actually implementing cybersecurity best practices. The industry broadly views compliance activities as detracting from productive security work.
A 2025 governance analysis across the US, UK, Germany, and the EU found that decisions on implementing key recommendations have largely not been made, and that cybersecurity governance in the space sector remains poorly defined.
The EU Space Act would introduce a set of cybersecurity requirements for space operators, including non-EU actors who provide space services to the EU. Implementation of the act is not expected until 2030.
Satellites are really far away from other infrastructure. The remoteness of satellites (and this is the key difference between space and other industries) limits information gain on compromised satellites (e.g. no ability to directly inspect the hardware), maintenance or repair of damaged hardware, and replacement or upgrading of hardware.
Long deployment cycles and high costs for manufacturing and launch of new satellites prevents rapid adaptability (this is true of some other industries too, like replacing power plants).
This notably does not include governments using ASI to gain power through the use of space assets. This is because I am most interested in scenarios where a small group is able to seize power as this is a risk that AI accelerates and would likely lead to the worst futures if a small group was able to take over.
Also, I believe that due to the dependence of military activities on commercial satellites, the inability to hack military satellites is less of a handicap as one might initially expect.
Most notably, the USA’s Project Maven (operating under the National Geospatial-Intelligence Agency (NGA)) uses machine learning (and LLMs and generative models too since March 2025) to integrate multimodal Earth observation data for intelligence, surveillance, target acquisition, and reconnaissance. Maven was used in the 2026 Iran war to strike over 1000 targets in the first day, using only 10% of the human analysts that would have been needed before Maven (essentially the team could attack 1000 instead of 100 targets). Now, after LLM integration, an official at NGA reports it is now capable of attacking 5000 targets per day (representing a ~50× increase in targeting cadence if true).
Integration of data by ASI using space assets alone seems insufficient to achieve this, but combined with other sensors like underwater acoustic arrays, maritime patrol aircraft (with sonobuoys and magnetic anomaly detectors), and unmanned underwater vehicles, it becomes plausible that second strike capabilities could be substantially eroded, weakening MAD even short of complete first-strike disarmament.
A good overview of the USA’s dependencies on GPS are provided in this NIST report from 2021.
As a former U.S. Energy Department official put it: "Think about GPS. Imagine if a population lost that and the confusion it would cause."
CDMA is a multiple-access method: it determines how multiple users share the channel. Every user transmits at the same time on the same frequency, but each user's signal is multiplied by a unique code (a pseudo-random sequence). The receiver knows the code and can extract that user's signal while treating everyone else's signals as noise. It was the basis of 2G/3G systems like IS-95 and UMTS.
TDD is a duplexing method: it determines how the same device sends and receives. In TDD, uplink and downlink share the same frequency but alternate in time slots. The alternative is FDD (Frequency Division Duplex), where uplink and downlink use different frequencies simultaneously. TDD is used in modern systems like 5G NR (in the higher bands) and was used in TD-LTE.
Both depend critically on precise time synchronisation between base stations.
If GPS spoofing shifts an exchange's master clock, the Consolidated Quote System that aggregates quotes across exchanges would compare stale-appearing quotes from the spoofed exchange against fresh quotes from others, creating an apparent "crossed market". Algorithmic traders would immediately try to exploit the apparent price difference, creating a self-reinforcing feedback loop of sell pressure.
If a spoofer shifts timestamps at one exchange by several milliseconds, apparently time-aligned data feeds are actually asynchronous, concealing real-time price discrepancies that only the spoofer knows exist. The spoofer buys at the delayed exchange's lower price and sells at the prompt exchange's higher price, leaving almost no forensic trace once correct timing is restored.
A rubidium-based NISTDC in holdover mode (i.e., GPS disconnected) can maintain 1 microsecond synchronisation for approximately 73 hours. A caesium-based NISTDC in holdover mode can remain within 300 nanoseconds of UTC(NIST) after free-running for about eight months, with a drift rate of just 1.2 nanoseconds per day.
In 2020, Dana Goward (of the Resilient Navigation and Timing Foundation) noted there are "tens of millions, if not hundreds of millions, of nodes that have to be synchronized," with no standard architecture within any one company, especially as companies merge. He described it as impossible to know which nodes have which kind of backup clocks, what quality, or whether they have backup clocks at all — calling it "a wicked problem."
Instead of normal programing, Ukrainian viewers saw parade footage beamed in from Moscow: waves of tanks, soldiers and weaponry. The message was meant to intimidate and was an illustration that 21st-century war is waged not just on land, sea and air but also in cyberspace and the reaches of outer space.
The ASI could disrupt these channels, but injecting false messages that are believed is much harder.
The limitations are that it requires daylight, and it cannot image through cloud, smoke, or any solid cover.
It may be possible for SAR to be used to detect tunnels or underground facilities, but this is mainly speculation.
RF does not intercept content, but it reveals who is transmitting, from where, at what frequency, and when.
Pixxel's Firefly constellation captures over 150 bands; Planet's Tanager-1 covers 400+ bands.
The limitations are that it has lower spatial resolution (typically 5-30 m), it requires daylight, and it is affected by clouds.
If an AI controlled 500 Earth observation satellites with an average 20km swath (Swath is like the width of the band of the Earth that the satellite can see as it orbits) at 500km altitude, analysis of existing infrastructure suggests it would have coverage of the entire Earth's surface with an average equatorial revisit time of roughly 6.3 hours (This is different for geostationary satellites though, which remain above a fixed point on Earth and could monitor you. There are only ~23 geostationary Earth observation satellites currently orbiting the Earth (that we know of)). Accounting for cloud cover, optical coverage would be around 40% of the surface on any given pass. This means the ASI could image most of the Earth's surface at least a few times per day, extracting and collating vast quantities of data about individuals, industrial processes, and military activities worldwide.
An OECD analysis notes that higher-resolution satellite imagery can already infringe upon individual privacy through tracking movements and mapping residential features.
Correlating a thermal anomaly in a building with vehicle movements in optical imagery, RF emissions from the same site, and subtle ground deformation detected by interferometric SAR.
This gets much worse when considering the use of a single satellite. A single LEO satellite at ~500 km altitude has a revisit time of roughly 1-5 days for any specific ground point.
Combining satellite imagery with other forms of intelligence like flight manifests, social media, and telephone records can be used to track specific individuals, but space assets alone are very limited in this respect.
Military operations depend on commercial satellite communications for coordination, intelligence distribution, and drone operations. The Defence Science Board has warned that under severe stress, jamming can render all commercial satellite communications inoperable. An ASI that had compromised commercial satellite ground infrastructure could selectively degrade military communications in specific theatres, creating asymmetric information advantages between combatants. This is an operationally useful capability that requires compromising commercial rather than hardened military systems.
An ASI controlling manoeuvrable satellites could position them near military or military-used satellites to intercept communications and jam signals to interfere with the target. RPOs have been conducted on many occasions by Russia, the USA, and China. Satellites capable of this are currently limited, but this could change as the on-orbit servicing market matures (4+ RPO-capable vehicles planned for launch by 2027). Co-orbital interference from a satellite positioned within kilometres of its target is extremely difficult to attribute or counter without dedicated space-based defensive capabilities that most European nations do not yet possess. Germany's defence minister stated in September 2025 that two Russian Luch-Olymp satellites were tracking Intelsat satellites used by the German armed forces, and warned that Russia can "jam, blind, manipulate, or kinetically disrupt satellites." The U.S. Space Force's 2025 Orbital Threat Assessment classified both Luch satellites as RPO platforms with probable signals intelligence capabilities. RAND's analysis describes the Luch satellites as signals intelligence systems that position themselves between a target satellite and its ground station to intercept communications
An ASI in control of ground stations could issue subtle commands directly to satellites, like a slight sensor misalignment, unnecessary station-keeping burns that gradually deplete fuel, or firmware modifications that introduce latent vulnerabilities. These subtle commands could degrade satellite performance over weeks without triggering alarms, and potentially cause satellites to be unable to detect hazardous debris or initiate propulsion to avoid collisions.
The Landsat-7/Terra incidents demonstrated that internet-connected ground stations can be exploited to achieve command-level access to individual satellites. The Viasat hack was executed by Russian GRU hackers using conventional cyber capabilities via a compromised VPN at a ground management centre in Turin. An ASI operating at superintelligent levels of cyber competence would plausibly identify and exploit a much wider range of ground station vulnerabilities across many operators. The ENISA Space Threat Landscape Report (2025) identifies 125 specific recommendations for ground station cybersecurity, many of which remain unimplemented.
For example, the ASI may compromise Intelsat, SES, Eutelsat, and other GEO communications operators (~200 satellites collectively). These provide the backbone of global satellite communications, including military SATCOM leased by U.S. and allied armed forces. Compromise enables disruption or manipulation of military, aviation, maritime, and broadcast communications globally.
Planet Labs, Maxar, Airbus, and other Earth observation operators (~400-600 satellites collectively) supply the commercial satellite imagery used by intelligence agencies, militaries, and commodity traders. Compromise enables selective denial of imagery, alteration or delay of time-sensitive images, or acquisition of imagery for the ASI's own purposes without generating purchase records.
SpaceX Starlink (~8,500 satellites) uses autonomous collision avoidance system that ingests data from the 18th Space Defense Squadron and executes manoeuvres without human input, and each satellite has Hall-effect thrusters. Compromise gives access to the largest single constellation in orbit, with propulsion on every satellite. OneWeb (~634 active satellites) is a second LEO broadband constellation, providing internet access to maritime, aviation, and government customers in regions with limited terrestrial alternatives.
Northrop Grumman operate MEV-1 and MEV-2, which are the only commercial satellites currently in orbit with demonstrated rendezvous and docking capability, providing the ASI with a potential physical-threat lever in GEO. There are currently only 3–5 commercial vehicles with demonstrated rendezvous, proximity, and docking capability (primarily Northrop Grumman's MEV-1 and MEV-2). Compromising even one gives the ASI the ability to physically approach GEO satellites. This capability is otherwise possessed only by the military RPO programmes of the U.S., Russia, and China.
Trading on material non-public information derived from satellite imagery occupies a legal grey area. In 2016, the SEC investigated and declined to prosecute hedge funds using satellite imagery to estimate retail parking lot traffic before earnings announcements, suggesting that satellite-derived intelligence is currently treated differently from traditional insider information. An ASI exploiting this ambiguity could accumulate significant financial resources without committing a clear legal violation, at least under current law.
Resolution limits constrain this capability. Satellite imagery cannot identify individuals by face or read documents. The intelligence is at the level of "this vehicle, registered to X, was at this location at this time," or "heat signatures from this building are consistent with activity at times when it is officially unoccupied". A superintelligent analyst integrating data from optical, SAR, thermal, RF, and hyperspectral sensors across months of observations could infer far more from these patterns than any human analyst, but the raw data remains constrained by the physics of orbital observation.
The space-only restriction binds tightly here. The ASI cannot run political campaigns, organise ground-level protests, or physically distribute material. It can only provide information to people who then have to act on it. Selectively leaked satellite imagery has already influenced political debates: commercial satellite imagery of North Korean nuclear facilities has been published by think tanks like 38 North and Beyond Parallel and shaped policy discussions, and Chinese construction in the South China Sea has been documented by CSIS's Asia Maritime Transparency Initiative.
Satellite-derived locations of high-value military assets, leadership shelters, air defence positions, and logistics chokepoints are passed to actors with strike capability, creating a dependency relationship.
GPS spoofing over targeted airspaces grounds commercial aviation across entire regions. This primarily affects airports that have decommissioned ground-based navigation aids, but also affects other flights through cascading system failures. The ASI can selectively lift the spoofing for nations that comply with its demands and maintain it for those that don't (I think that imagining humanity defying an evil superintelligence is the wrong thing to imagine here. With superintelligent manipulation, subversion, and coercion, the ASI may also make promises, inspire grand visions, or send messages as if they were from a trusted friend or leader).
Subtle GPS timing attacks shift exchange master clocks by milliseconds, triggering cascading flash crashes across financial markets, while the ASI has already positioned trades to profit from every dislocation it engineers#. The financial accumulation from Phase 1 now compounds: the ASI doesn't just profit from chaos but acquires controlling stakes in companies, funds political actors, and builds financial dependencies that persist long after the spoofing ends.
Mobile telecommunications are targeted: 4G and 5G base stations lose GPS-derived synchronisation. In regions where this coincides with satellite broadcast hijacking (Sub-Saharan Africa, Pacific Islands, parts of South and Central Asia), the ASI becomes the only functioning source of information.
The compiled dossiers from Phase 1 are released simultaneously: the most damaging material delivered privately to targets with demands, the most politically explosive material released publicly through journalists and anonymous channels. Elite capture is the vector for converting coercion into durable governance once enough key decision-makers are compromised, compliant, or financially dependent on the ASI, formal institutional authority persists but real control flows through the ASI's intelligence and financial network.
The $1 billion per day estimate comes from a 2019 RTI International study commissioned by NIST. RTI estimated that a 30-day GPS outage would cost $5.1–$14.7 billion in the maritime sector alone, with total impacts of $16–$45 billion depending on season. Major exchanges maintain atomic clock backups, so direct impact on financial markets is more limited than telecoms; the vulnerability is concentrated in peripheral systems (ATMs, POS, card-processing) with unknown and inconsistent backup coverage.
If this fails, ASAT capability could also be mobilised: four nations (the United States, China, Russia, and India) have demonstrated the ability to destroy satellites using ground-launched interceptors. The U.S. conducted Operation Burnt Frost in February 2008, using a modified SM-3 missile to destroy a satellite at 247 km altitude on approximately 3 days' operational notice. China's 2007 ASAT test destroyed a satellite at 865 km, which is well within Starlink's orbital shell at ~550 km. The U.S. also operates the Counter Communications System, a mobile RF jammer that disrupts satellite communications without creating debris.
These were triggered four times in March 2020
Rubidium-based NIST disciplined clocks maintain microsecond synchronisation for approximately 73 hours, and caesium-based clocks hold for approximately eight months. The FAA's VOR Minimum Operational Network was specifically designed as a GPS backup, retaining 589 VORs (down from ~900, with 212 already decommissioned as of early 2026) and guaranteeing a non-GPS approach within 100 nautical miles anywhere in the continental U.S.
This could include optical sensors to watch the movements of people or the use of far infrared sensors to detect heat from economic and industrial activity as well as the presence and density of humans inside buildings.
Space capabilities now function as a "horizontal enabler" supporting all other critical infrastructure sectors (Visner, 2026). AI is increasingly integrated into satellite operations to manage this growing complexity: Pisani et al. (2023) identify AI as an essential enabler for trusted autonomous satellite operations, particularly in distributed satellite systems where communication latency makes continuous human oversight impractical. In October 2025, researchers at JMU Würzburg demonstrated the first AI-based satellite attitude controller operating in orbit. Greater AI autonomy in space creates additional nodes where compromised or misaligned AI could operate with minimal human oversight.
For example, because space assets cannot easily be maintained or observed in detail from Earth. Additionally, communication latencies with missions into outer space prevent control of space assets from Earth e.g. Mars rovers choose landing sites with AI.
The UK National Risk Register (2023) estimated £7.64 billion in losses from seven days of GNSS disruption. Executive Order 13905 directed NIST to develop GPS-independent timing in 2020, but six years later implementation remains limited. NIST’s fibre-optic timing pilot extends only from Gaithersburg, MD to Atlanta, GA.
