Mirror Bio and the ARC AP-04
Understanding the threat
The key threat we’re concerned about is mirror biology.
Others have done a much better job than I could do of summarizing the findings and explaining the danger of mirror biology, and the team behind the report have also created a short summary article. However, these summaries do not go into much detail about possible defenses, and the technical report doesn’t discuss the possibility of shelters. To be clear, shelters are in no way a desired outcome: they would be a desperate, last-minute effort to save what we could as the biosphere turned hostile.
Although the threat framing here focuses on mirror biology, some have argued that this narrow focus might be counterproductive if it diverts attention from broader synthetic biology risks or other extreme possibilities, like alien biosphere colonization. Mirror bacteria can indeed be conceptualized as one specific instance among multiple engineered pathogens, so the defensive measures proposed here could apply to all scenarios involving dangerous particles in the atmosphere. Nevertheless, mirror biology remains especially concerning for two reasons: it could evade many immunological recognition pathways (though not necessarily all) and, if it reshapes the biosphere, it might prove far harder to contain than a typical pathogen with conventional biochemistry.
What’s needed for shelter design is to understand environmental concentrations of the threat. Unfortunately, as the report explains, we’re still quite uncertain how a mirror biology catastrophe would play out. This means we will have a hard time guessing about what environmental concentrations we have to defend against. With uncertainty, it becomes necessary to create high levels of protection. The strategy we employ below is to use upper levels of currently observed “normal” microbial concentrations and then add a safety factor on top of that to represent the risk that, without natural ‘predators’, mirror bacteria might temporarily become overwhelmingly common.
The highest atmospheric concentrations we’ve identified are in connection with dust storms. This makes intuitive sense: soil is some of the most microbially dense widespread material we know of and high winds will bring large amounts of soil and dust into the air. We discuss concentrations in units of Colony Forming Units per cubic meter, CFU/m3.The highest CFU counts recorded during dust events are around 10^7 CFUs/m3.
This is not as conservative an estimate as we would prefer, again due to the uncertainty about exactly how a mirror biology catastrophe would unfold. Mirror organisms would inevitably interact with the complex environment in a multitude of hard-to-predict ways, and if significant vegetation dies this could potentially lead to much higher erosion rates and more frequent dust storms in areas that have not previously experienced significant dust storms. Still, we think 10^7 CFU/m3 is still a generous upper bound when considered over the multiple years the shelter would be in operation.
How would we defend against a concentration of 10^7 CFU/m3? Air filters are the obvious choice, but how much filtration would we need? The requirement for removal is to not let a single particle into the lungs or digestive tracts of the inhabitants, because we want to conservatively assume that if this happens the microbe will reproduce, killing the initial host and any other shelter inhabitants.
In microbiology and related fields, due to the extreme numbers of microorganisms as well as their exponential growth, one uses logarithms to talk about sterilization. Reduction by 90% of a microorganism is a 1 log reduction, 99% is 2 log and so on: the logs can be thought of as “counting” the numbers of 9s in the percentage efficiency number. Now, no number of log reductions will give us certainty that no CFU makes its way to the inside; we can only talk in probabilistic terms. Therefore, let us start with the requirement that we want a 1% chance or less that a single CFU is inhaled by an inhabitant.
Consider a shelter designed to protect four people for one year. Each person needs at most approximately 20m3/day (see e.g. table 6-5 here) of fresh air, but assume 40m3/day to be conservative. This would require an air intake of around 160m3/day, or approaching 10^5 m3 over a year. At our target average atmospheric concentration of 10^7 CFU/m3 our filtration system will be faced with 10^12 CFUs. To have just a 1% chance of passing a CFU through we would need it to pass fewer than one in 10^14 CFUs, a 14-log reduction. This is a staggering reduction, but as we discuss below we think this is possible with sequential filtering.
Water concentrations are similarly hard to estimate, and in current shelter work we have accounted for consistent, extreme levels. Note that for water, concentration numbers can be much, much higher than for air (at some point the definition of “water” is cast in doubt - it could be mostly microbes mixed with a bit of water!). For example, in water just downstream of large amounts of feces or decomposing carcasses we would expect to see something in the range of up towards, and perhaps sometimes above 10^8 CFU/ml. The latter scenario could be a common occurrence in a worst-case mirror biology catastrophe. With heat sterilization, a robust method, we can assume one can sterilize to 10 logs, probably even quite a bit more. But this would be insufficient for extremely polluted water over longer time periods. Therefore, we would recommend sourcing water from an old aquifer - these can take more than 100 years to receive significant intrusion from the surface and on a per liter basis, especially over the long-term, such clean water supply is extremely cost effective compared to other methods of delivering safe water.
Even if we built a shelter that could keep out this level of environmental hazard, we think this is unlikely to be a scenario where humanity can simply stay put and wait for the problem to go away. We see shelters meeting these requirements as only one component of a larger response, allowing more people to survive to a time when, through efforts elsewhere, it’s possible to live outside these shelters again.
One question that is probably high on people’s minds and that is also very relevant to shelter work: Is it likely that we will as a global society develop dangerous mirror biology science? To this I can only say I really hope that we can keep a lid on this, but I want us to be prepared in case that’s not how it goes.
Details on the shelters
The current shelter design is fundamentally uncomplicated: A positively pressurized plastic “bubble” supplied by serially filtered air. That extreme levels of protection can be achieved with simple and relatively affordable protection makes this solution attractive.
These shelters are a direct descendant of a lot of different strands of previous shelter work. They build on the civilian nuclear shelters in Northern Europe, continuity of government bunkers in the US and Russia, Collective Protection Units used in the military, and concepts of civilizational shelters or refuges discussed on this forum by various people since 2014. However, around 2021 there was an increase in action around this idea. It is unclear to me exactly what drove this increased interest: it could have been the seeming availability of funding for catastrophe prevention, the gradually rising prospect of a mirror biology catastrophe, or something else. This post describes work that directly built on that increased activity, encouraged by previous suggestions that shelters be pursued, using previous work as input and announced in a previous post I wrote declaring the commencement of my work on the topic.
The shelters were conceptualized as an answer to the following question: what would be the absolutely cheapest way to construct a space that had 14-log protection in terms of atmospheric aerosols? When the question is phrased this way a solution presented itself: serial air filters supplying a positively pressurized plastic bubble tent, inside a larger existing structure for protection from the elements.
While the concept of a positively pressurized shelter isn’t new, we’re not aware of earlier work that uses serial filters. Moreover, this concept of a shelter is extremely minimal, which has two additional benefits:
Very cost effective
Possibility of rapid scaling of production
Advanced serial filtration has demonstrated exceptional performance in controlled environments. During the Cold War there were plants that generated plutonium dust and needed to vent dust-containing air to the environment. Due to concerns about radioactive pollution, air was passed through a series of HEPA (protection factor of 2000 which is 99.95% efficient) filters and the efficacy of this treatment was finally tested at the Los Alamos lab that demonstrated an average of 12-log performance and a worst-case performance of 10-log. We are therefore fairly certain that this performance can be extended to 14-log and perhaps even higher.
For the positive pressure, no similar empirical experiments at the required level of performance have been found. But talking to an engineering professor in cleanroom technology who has investigated contaminant transport into cleanrooms, they thought it impossible for even a single particle to enter a positively pressurized space through the space envelope. Moreover, calculations were performed on diffusion speeds and likelihoods based on established physics and these similarly showed that practically speaking, the chance of a particle entering “against the flow” through a 0.5mm wide and 2mm long hole was, for all intents and purposes zero.
Wind pressure is a critical consideration in shelter deployment. Simple calculations with Bernoulli’s equation show that one can quickly get pressures of more than 100 Pa with wind gusts that appear with some frequency in most locations. If the pressure generated by wind exceeds the pressure differential from the inside to the outside, there is a significant risk that outside aerosols might be pushed inside. This is why these shelters are envisioned being deployed inside a larger protective structure. Due to the inflatable plastic structure, there are few requirements on such spaces and they can be anything from garages and large living rooms to farm buildings and warehouses.
One might be concerned about real-world “points of failure” like leaks through gaskets, tears in membranes, or mechanical/electrical breakdowns that might not be as well-controlled as in a laboratory cleanroom. While positive pressure helps ensure that any small puncture leaks air outward rather than letting contaminants in, larger tears or equipment failures could compromise the shelter. Cumulative stressors—like regular use, environmental damage, or accidental collisions—could cause small holes to expand. Therefore, we plan to leverage self-limiting, tear-resistant fabrics, multi-layered designs, and redundancy for critical components (like fans and power supplies). Survival under as extreme circumstances as these shelters are designed for cannot be guaranteed. However, real-world testing and careful material selection and robust construction methods will make these shelters robust under a wide range of actual use scenarios.
While the main concept of these mirror bio shelters is a smaller positively pressurized space supplied by serially filtered air, there are more components needed for long-term survival.
Waste is ejected via a specially designed waste system that similarly to BSL 4 labs do not let potentially contaminated air go “the wrong way” (e.g. bubbles or biofilm going from the dirty to the clean side of the shelter).
Entry and exit is perhaps the most vulnerable part of the intervention. The only empirically tested, high-log decontamination found was for germ-free laboratory animals (“gnotobiotics”). Here, Vaporized Hydrogen Peroxide (VHP) is used to decontaminate, and animals are transferred between cages by VHP sterilized and air purged tunnels. One can imagine people being transferred in a similar fashion especially between shelters and vehicles in order to facilitate a functioning society.
Above, the following items have been covered:
Air supply
Protective structure
Water supply
Waste handling
Decontamination/airlock
In addition, the following items are likely required:
Protective gear for habitation transfer and outside missions
Power
Food
Less critical but important items like bedding, exercise equipment, etc.
On protective gear, the highest protection factor gear found has been >50,000 protection factor which is 4-5 log of protection. Note that this is far short of the required 14 log for the protection. Some of this gap can be bridged by limiting the amount of time spent outside (if needing to survive for only 1 hour, the required log reduction would be “only” 7 log). Also, if combining a suit with protective tunnels to transfer personnel between habitation and transportation, it might be that the tunnel + suit will offer sufficient protection. Moreover, these suits will be supplied by stored, compressed air so the tunnels could be filled with VHP, further increasing the log reduction.
For power, it is hoped that the government will protect the utility workers so that power will be available via the grid. But in case one would like to prepare for the eventuality that this fails, or even to have protection against interruptions, an off-grid system might be good. The most cost effective set-up will depend on geography. In areas with sufficient sunshine during winter, solar and batteries will provide the main bulk of power while a propane generator will provide power during any prolonged periods of cloud cover. Note that the most costly components of an off-grid system (solar and batteries) can be used during regular periods to offset utility bills and therefore partially (or in special cases fully!) pays for itself.
For food, it is fortunate that the Church of the Latter Day Saints has been developing cost effective ways for long-term storage of food. There is some uncertainty about especially vitamins and oxidation of fats, but it is hoped that refrigeration will go some way to solve this issue. In any case, based on a growing base of information from space missions and Mars analogues, it seems very wise to make a small investment in an ability to grow plants indoors. Organic waste will be ample, and there will be water. As such, at least for some time, it should be possible to at least grow some foods that could help alleviate especially problems around vitamin deficiencies.
Other items are important too, even though they might not directly relate to the rule of 3. Long durations of isolation places very high burdens on people and the lockdowns many experience during COVID was quite benign compared to being sealed in bio shelters for months, if not years. Luckily, Tereza Flidrova has done excellent work on what is needed to increase the likelihood that significant psychological problems do not happen and the shelter design should heed as much of this advice as possible. Luckily, due to the flexible and low cost material, many such design aspects can quite easily be accommodated at only modest increases in cost of production.
The first version of the shelter structures, “plug-and-play” ready are expected to retail for $39k. The structure would include the following components:
>14 log serial filters along with certified low-leakage ductwork - high-quality, industrial grade components
Bubble - high quality ones that have been deployed without failure for years in climates as diverse as the Wadi Rum desert and Iceland (material and construction used in “bubble hotels” meaning they also comply with fire safety standards)
Sterile water supply (includes heat sterilization but excludes ground well)
Waste system
Note that the earlier $10k/person number does not include anything but material costs. This is because it is unclear how, in a “war time mobilization” by the government to make as many units as possible in the early days of a crisis, how the cost of manufacturing etc. will be accounted for. The design might even evolve to be simple enough for people to make such shelters by themselves out of commonly found and varied plastic materials and HEPA filters repurposed from other uses. This would indeed be a highly desired outcome and I commit to always optimize for societal goods. For example, if a part of the shelter was to be protected by IP held by the company set up, I would do all I could to let individual households use the design freely. If the company is successful I will also be careful about what investors to bring on board as a disastrous outcome would be one where there is an unfolding mirror bio catastrophe but owners of patents refuse to share the design with the wider world in order to maximize profit. An ideal outcome would be to have a high level of governance and transparency which is why one board member has been chosen specifically for this purpose.
Beyond this, the following would further increase the robustness of these shelters (note that power and food can be consumed and as such might at least partially “pay for itself”):
Power - In a sunny location with climate that does not require air conditioning during crisis use this is estimated to be ~$12,000 (source: my own comprehensive timer series modelling of solar radiation from Spain’s Mediterranean coast)
Food is estimated to cost $1,200/person/year (source: from an expert in food storage which is again based on e.g. current prices on the LDS store)
Additionally comes furniture, lighting and decoration
Lastly, in order to exit the shelter during low atmospheric concentrations, the following would be needed in addition:
BSL 4 reusable suit - ~$2,000 (source: both online cost indications but also checked with GPT)
VHP generator - ~$5,000 (source: based on various online cost quotes)
Power might be more expensive, and in some cases significantly more expensive in locations with less sunshine, and especially if heating beyond electric blankets are needed. However, the suggestion with the intervention, if cost is a concern, is that gov’ts will focus deployment in locations where the climatic conditions are more benign. Luckily, this often happens to coincide with large population centres as people often chose to settle where the weather is better (e.g. California, Southern Europe).
Lastly, as the currently designed units planned for immediate sale is based on comfortable bubble hotel construction and design, it is imagined that in certain jurisdictions, these units can even be used during “peace time”, when there is no imminent crisis. For example, they could be put up on a lawn to provide space for guests or teenagers. Or if one has a remote piece of land, as a weekend getaway. As such, the hope is that this will sufficiently increase the attractiveness of these units so that a number of them are actually deployed, marking real-world progress on an “end-to-end” x-risk intervention: Advisors have speculated that if a sufficient number of these units are deployed, this might have already decreased existential risk by some amount, especially if we can get some distance beyond ~100 units over a not-too-large geographical area. And given the relatively modest philanthropic funding of this project to date, this effort might hint at a cost effective, “end-to-end” x-risk reduction, especially if gov’ts take the necessary steps to prepare to produce these units at scale. Thus, the ideal scenario is one where governments are ready to produce thousands of units so not only a minimum viable population survives, but enough people to carry on the most critical, welfare-generating parts of our societies.
The road ahead
At this point it might be worth revisiting the epistemic status of the topic of how these shelters would actually be used in a mirror biology catastrophe. Put succinctly, the epistemic certainty drops significantly when speculating on the road ahead. So far, these units seem to physically offer significant protection and they might be tolerable from an inhabitant well-being perspective although larger units would be desired. But both because there is inherent uncertainty about exactly which mirror pathogen would be the concern, as well as how any mirror pathogen would interact with the environment it is really hard to say what surviving such a catastrophe looks like. For example, might there actually be periods with sufficiently low atmospheric concentrations so that people can be outside with only 2-3 log protective PPE?
Scaling to millions of units, training people to use suits and shelters properly, and dealing with real-world complexities (like suit maintenance, vehicle transfers, and power-grid reliability) could pose challenges. We acknowledge these hurdles and do not claim a guarantee of everyone’s survival under every circumstance. Rather, we aim to offer a significant probability boost in particular high-risk scenarios—similar in spirit to nuclear bunkers or lifeboats, which are not failproof but can save many lives when properly built and utilized.
Still, one step seems clear: We need to take these shelter plans from paper to reality, and start producing, testing and improving on these shelters.
There are some “binary” thresholds in terms of the number of units deployed in a crisis:
At deployment levels sufficient to ensure the survival of a Minimum Viable Population (MVP), there is at least some chance that a small deployment (likely 100s of units) will be the difference between extinction and survival of intelligent life in the universe. However, with such small numbers it is unclear what the post-catastrophe game plan looks like, how habitable the planet will be, etc. Most likely, this would work in the subset of scenarios where for some reason there is a high number of mirror microbes for the first few months but that after this the levels subside to something more manageable (e.g. requiring “only” 2-5 log protection during small periods of the year).
At a deployment level that includes enough skilled labor of the right types, not only will a higher chance of long-term survival be possible, but one might even have made possible the continuation of high welfare communities even if the biosphere is permanently altered to something significantly more hostile to human life. This could enable the continuation of liberal democracies or similarly welfare-supporting forms of society.
Lastly, at deployment numbers in between the two scenarios above, there is a slowly rising chance of survival and the continuation of societies that support human flourishing. But this is far less than linear: For the first additional units beyond a few hundred - things do not look much better at all. E.g. doubling the number above what is required for MVP survival does not at all double the chances of “rebuild”, especially of something like liberal democracies. But at 10% units less than the deployment level that right away supports continuation of liberal democracies, one probably has almost the same chance of rebuilding as with “all” the units.
This text was initially posted on the LW forum and has been adopted for this website.