ASI Safety Lab

Political Aspects of ASI Safety

We are living in politically divided nations and in a divided and segregated world. All nations are sovereign entities. They have supreme control over their territory, and therefore the final word on their governing laws. Many countries, the middle and superpowers, are starting to get very concerned about their national cyber defense capabilities.

Unfortunately, there is a serious lack of imagination among politicians and world leaders about what AI, ASI could actually mean for them. Some politicians and entire countries think they could become winners. I have serious doubts about that.

AI/ASI has some cultural impact on the conversation and on our expectation of what it could mean. TV series and movies reflect these views and fears. I don’t want to become a movie or TV series critic, but some plots may have some grain of truth in them — However, politicians are playing public sentiments both ways. That means some of them won’t necessarily be opposed to creating a brighter picture of what ASI could mean to us and that they will support measures to make it safer instead to stick with the doom and gloom.

Additionally, we need politics to draw some common-sense lines on what governments are allowed to do in their territory. I don’t think that mankind could allow AI/ASI to corrupt governments and use their corresponding territory as a protected retreat area. We could not allow that ASI gathered covertly resources to have other governments fall like dominoes.

Finally, I want to discuss some common-sense regulation and legislation.


Although countries have agreed upon international treaties and rules, there is still fierce competition to undercut some countries’ even common-sense regulations. If there is a chance that no regulation could lead to some, even a minor, advantage for even a small group, then there is a good chance that someone will use this opportunity. Many countries see AI as a tool that could change their standing and position in the updated world order. Some political leaders are well aware that the country that has AI/ASI first may lead the world potentially irreversibly and unchallenged. It is conceivable that some middle power could become the new world-/super-power — while the military of the current superpower would be turned inert.

A digitally decapitated superpower … this sounds impossible, but it’s not. Every military power is (literally) based on logistics and complex technology. Modern logistics requires a reliable and defensible computer and communication infrastructure. If the software of this infrastructure is sabotaged in millions of different places or occasions by an ASI, then this superpower is finished. Right now, this is not doable, because hacking is a human skill. But soon, it is conceivable that ASI could take control over computer systems in a totally covert manner — so that this superpower would not even be aware that they are already finished — but the ASI could be sure about it and would camouflage its presence and capabilities. Remember: the Russian SolarWind hack was “undetected for months”. There is a deeper reason behind this: Complex technology can’t be safe or can’t be protected reliably. Every IT system is based on a von Neumann architecture — and that means regular data can be turned into malicious code (scripts or apps). There is simply too much we must watch so that we can be sure that a machine is not being taken over. As long as the adversaries are humans, we would face only a few incidences, but when hacking/cracking becomes a tool of a smart ASI, then we are at its mercy or at the mercy of the entity that is controlling that AI/ASI.

ASI Safety is currently seen as part of Cyber-Defense or Cyber-Warfare — if at all. Countries are investing billions in upgrading their IT infrastructure to be prepared. But these investments are a small patch, and they are being forgotten literally the next week or next month because there will be a new cyber-threat. Unfortunately, this will continue, because the current type of cyber-defense is a reliable source of revenue and income. So far, no business has had the guts to suggest or provide a solution. Why? Has nobody asked to get a more secure system? Most in charge would argue that the threat might not be significant or essential enough: as long as cyber-attacks are done by a few highly skilled humans, other humans are hired to take care of that problem with labor-intensive tools.

Right now, we have only trusted devices — and these devices could become untrustworthy. Same with governments, they are hiring trusted people — but some become traitors. We need to do better with technology. We need trustworthy computer systems — something that can’t be turned into something that is controllable by an adversary. If we truly know based on first principles that a computer system is trustworthy, then it would literally take a violation of the laws of nature to be taken over by an adversary. I have a posting: “Safety based on First Principles” that is discussing this issue and is describing what needs to happen so that we have not only trusted but trustworthy computer systems.

I believe, most nations will understand that they will need trustworthy computer systems, and they will demand from their IT supplier solutions that would not only be safe against human attackers but secure against any attacker (including ASI), because a few, seemingly minor, changes in the computer architecture could significantly change the cyber-defense landscape. To name only a few: Harvard-Mode for CPUs, Storage Watchdog for executable files and Port Certificates for Firewalls are relatively easy to implement but significant and huge in their impact.

IT Ecosystem

ASI-Safety is more than nations’ cyber-warfare or defense capabilities. We need to understand that the entire IT ecosystem, i.e. all electronic devices with CPU, RAM, storage media, and networking capabilities are a possible battleground for ASI. At the time that we detect ASI could be a threat to humans (to our life or our freedom), or to the foundation of our technological civilization (our infrastructure, our money system or e-Commerce infrastructure, etc.) then ASI may not be omnipresent on all different device-types yet; but can know, and we should better not make assumptions on what ASI can or can’t do prior to an ASI conflict event. Being prepared means that we need to be prepared for the full spectrum of IT vulnerabilities.

The lifecycle of some IT products is helping the ASI Safety effort. New phones are being bought every 2 or 3 years. Computers are usually replaced every 3 to 5 years. Unfortunately, CDs and DVDs have a life expectancy of 100-200 years, but others say they are only reliable for the next 25 or 30 years. Memory cards are not supported by new devices anymore. Thumb drives are getting lost or broken after a few years. There are 3 aspects to know: (1) there are collectors of legacy devices; they are usually not using them, but ASI was potentially able to infect them; (2) some users are keeping old machine until they die (which explain why we still have PCs with Windows XP); (3) there are users with little or no resources, but they got donated very old, potentially devices; so these devices wouldn’t show up in the context of new machines with ASI-Safety technology — and without additional effort, ASI might actually survive in these very old legacy systems, in the hand of poor people.

In a world in which every user has multiple devices, some are talking of 50 or more devices per person (on average), keeping them all loaded with energy is certainly not a task that motivates users to buy more gadgets. Therefore, it can be expected that new devices will either wirelessly loaded or will be able to harness energy out of the environment. So having many devices entering the market and not knowing how long they will last is a problem.

On the other side, we don’t know how fast AGI/ASI will become available. Additionally, we don’t know by when we should have a Kill-ASI Switch available. I would guess around 2030, we should have some level of ASI-Safety. There is

National Sovereignty and ASI

Countries agree and accept in international treaties that international law is superior to their corresponding national law. As part of their signing, they often turn the treaty officially into national law via resolutions in their corresponding parliaments.

Political leaders in all countries must understand that they are uniquely vulnerable to open or covert attacks by ASI. The attack could also take the form of bribery or a blatant blackmailing attempt. ASI will understand the leader’s weaknesses than he knows it by himself. ASI could frame a strong-minded leader (that he committed some crime — the ASI may even fabricate the evidence) and then push a weaker candidate to the front without the social environment would even recognize that. Being a leader or even an adviser or any person with any power (formal or informal) is in danger to be manipulated, tricked, or forced to do something that would be in the interest of ASI, but not necessarily good for the country and/or its citizens. ASI may change the facts communicated to the leader, or change systematically advisers or undermine the trust in his team.

But what if an ASI finds a leader of a country who is a willing traitor who sells out the freedom of its citizens for favors to him personally. We all should be concerned that some people, including leaders, will believe ASI’s promise of delivering special favors. These are devil pacts, and it is very questionable if a devil is honoring whatever it promises. On the contrary, ASI would not feel bound to human honor, integrity, honesty, or morality. A lie or a false promise is a simple tool and nobody who could tell a tale would potentially survive to tell that tale. And even if someone would survive, it is more likely that a leader, i.e. a person with power, could become a stooge, or even a prisoner of ASI.

The world community may need to agree on methods and procedures on how they decide if and when an ASI needs to be killed. The world community must have the right to go against compromised leaders in order to protect themselves as the next fallen dominoes and to protect mankind as a whole.

Besides the threats ASI could face for the political and legislative leadership, countries are responsible for what ASI can or can’t do, or how people, groups, or companies can interact with ASI, and what limits and/or rules/laws both sides (ASI and humans) must follow.

It is likely that ASI will be bound to different rules and/or laws. It might be required to define some minimum standards that countries need to commit so that ASI is not getting out of control in territories with less strict rules.

Common Sense Regulation/Legislation

Most powerful technologies have safety measures against accidents. Actors (individuals, companies, and even countries) are kept accountable and liable for their actions or in-actions. Classical software is based on a clear cause/effect relationship. If something went wrong, someone failed, and therefore someone is subject to product liability laws. With deep learning, we have systems with training and test data that are being prepared. However, we often don’t know how these systems behave, but this should not be used to limit the responsibility and accountability for someone’s product that is using prepared training data. For some time, we even need to extend the liability for products that are learning from the environment, and nobody knows what data it uses. Even in these cases, we can’t let manufacturers off the hook for their liability.

Additionally, there should be no doubt about legacy systems. If an old system or old storage devices or media can’t be cleaned reliably or their reinfection with ASI can’t be prevented, then all these devices and storage media must be destroyed. People must potentially be made liable and punished if they fail to follow the destruction order.

Unfortunately, some people will not only defy these orders but potentially collude or conspire with ASI. A conflict with ASI is a war situation. People who help the enemy are traitors and they must be treated accordingly.

Most importantly, I believe, will be the commitment of countries or economic zones to make ASI Safety default for new IT devices, like the seatbelt. And soon after, help to make tagging of old legacy devices mandatory as well. The legislator must potentially demand that certain software providers include technology that checks for the availability of tools that would help to deactivate legacy devices in case of a Kill-ASI event


I hope that ASI Safety technology is being developed that will allow the identification of individual ASI instances. In that case, Kill-ASI should then be more about law-enforcement for violating rules than about killing indiscriminately all ASI instances. Additionally, I hope we have enough technology to detect in principle rule violations committed by ASI without giving ASI a heads-up when an event is being analyzed based on a randomized inspection process.