AI ethics: law, medicine, education, and war

Google’s AI overviews are accurate 91% of the time... 91% sounds reassuring until you apply it to 5 trillion annual searches because the remaining 9% translates to roughly 57 million wrong answers every hour... the consequences of that 9% would be real if a commercial airline had a 91% success rate. We wouldn’t call it aviation. We’d call it an extreme sport…
– House of El

We have skirted themes around AI and robots several times in the past ten years – touching on sex and robots, self-driving cars and what these reveal about cultural/moral relativism, AI governance,1 autonomy and the “rights” of agential robots, AI therapy, etc. These topics were, and still are, more or less speculative, science fictiony – gist for armchair philosophy. The topics we address this time are looming, if not already present, realities.

The world has changed in the last few years. It is likely that more literature has been published on the topic of this article since the first version (early 2020) than in the entire time before.
– Vincent C. Müller, author of the SEP article on “AI Ethics” (rev. March 2026).

The full writeup is here.

Philosophical ethics is rapidly coming to terms with the “AI revolution” as it once had with the “industrial revolution.”2 We are now in a position to survey the actual emerging philosophically-laden landscape surrounding AI/robotics and our interactions with it. The AI specific area of ethics, both theoretical and applied, is maturing rapidly as problems are becoming clearer to scholars across many fields. Ethics is the upstream conversation and source of all decisions about what should be done or thought – at least for entities (human or otherwise) who fancy themselves having agency, i.e., able to decide and act based on rational consideration.

Al and law
An area downstream of ethics is the philosophy of law. How the law is to apply to autonomous AI, if it really is autonomous, is being decided as this is being written.

A German court just ruled that Google’s AI Overviews are not neutral search results, but they are Google’s own words. And in the same week, a US federal judge sanctioned lawyers on both sides of a case for submitting AI-generated legal citations to cases that did not exist.
– House of El

House of El, computer scientist and geopolitical data analyst, suggests this may be the beginning of a comeuppance long overdue for those who claim that AI autonomy is a shield from responsibility.

Is AI autonomy real? If it is, we need to start burning down data centers as millions of cows were during the mad cow disease scare. If it is not, as is more likely, some people behind the claim that it is may require prosecution for the same reasons we regulate (or ought to regulate) drug dealers. This, because we recognize that those who facilitate irresponsibility in others should participate in the penal consequences. You cannot escape responsibility, let alone profit from, a situation you created and claim immunity should that situation cause harm. You cannot have it both ways: claim it did it on its own and take material credit for having created it.

The logic of these courts is, as EL says, “highly portable.” It does not quite determine the law in Germany, the U.S., or anywhere else, yet. But the reasoning is pristine and the law everywhere is under a defeasible obligation to appear rational wherever human cussedness permits.

On this assumption, the AI shit may be about to hit the fan…

But so far these are just civil cases, nothing redistribution of money couldn’t palliate. The stakes can be higher...

AI and medicine
Want a disease to go with that vaccine?... We got one.

According to the very recently published (May 2026) 500 page The Biosecurity Handbook: Biological Security in the AI Era by Bryan Tegomoh, MD, MPH, gain of function (GoF) research:

...involves genetically altering organisms to give their gene products new or enhanced capabilities. In the biosecurity context, GoF specifically refers to experiments that enhance a pathogen’s:

• Transmissibility: Making it spread more easily (e.g., airborne transmission)
• Virulence: Increasing disease severity or lethality
• Host range: Enabling infection of new species (e.g., making avian virus infect mammals).3

I asked Google’s AI on 17th June 2026 about AI’s role in gain of function research. (I specify the date and AI because if you ask a different AI or on a different day, or even time of day, you may get a slightly different response.) Google’s AI did very helpfully point me to The Biosecurity Handbook the latest, most thorough treatment of the subject. But mostly it boasted of AI’s contribution. (It did sound like a “boast,” or a very smart but naive child uncritically excited about a wonderful development.) Google’s words:

Artificial intelligence is increasingly transforming virology and synthetic biology. Rather than directly performing experiments, AI accelerates Gain-of-Function (GoF) and similar research by predicting how genetic mutations affect a pathogen’s transmissibility, virulence, or host range.

The role of AI in this domain spans several specific capabilities:

• Predictive Modeling: AI models are utilized to predict how specific mutations alter viral proteins, helping researchers simulate protein folding and foresee how a virus might evade human antibodies or jump to new species.
• Directed Evolution and Red Teaming: Machine learning algorithms can systematically test millions of viral variants to identify ones with desired traits. This computational “red-teaming”4 is functionally analogous to the directed evolution seen in GoF research.
• Hypothesis Generation: AI systems act as “co-scientists,” suggesting testable experiments and identifying genetic targets that could sharpen the understanding of viral replication and pathogenesis.
• Accelerated Countermeasure Development: By understanding and even virtually simulating these mutations, AI is highly valued for predicting emerging threats and rapidly designing targeted vaccines or therapeutics before an outbreak spreads.

As artless as a child, AI may insouciantly incriminate itself. It will hand over the evidence, the good and the bad, seemingly unaware that a bold face conjunction of the two requires a synthesis. It did mention biosecurity concerns and offered the GoF link to The Biosecurity Handbook, which was prepared for training professionals in both biological research and medical AI.5 The handbook states near the beginning:

The same AI that accelerates drug discovery can accelerate the design of dangerous pathogens. Biology is getting easier to engineer, and governance has not caught up…

Gain-of-Function Research: Science, Risk, and Governance

In 2012, two research groups demonstrated that just five mutations could make H5N1 avian influenza airborne-transmissible between mammals. The same virus that kills 50-60% of human cases but rarely spreads person-to-person could become a respiratory pathogen. The studies sparked immediate controversy. Critics called it a recipe for a pandemic. Defenders argued the knowledge was essential for surveillance and vaccine development. More than a decade later, the scientific community remains divided on whether we can safely create pandemic threats to prevent them. [emphasis added]

Getting ahead of ourselves
A few weeks ago, Tulsi Gabbard, as she was about to leave her post as Director of the Office of National Intelligence, released classified documents detailing U.S. government funding of biological research labs in 30 countries that engage in gain-of-function research, some of it dating back to the George W. Bush administration but implicating every U.S. administration and government since, regardless of political party. Famously, Dr. Anthony Fauci (among other government medical authorities) denied he had any role in these activities. The documents suggest otherwise.

Why in 30 foreign countries? Why classified?

Many U.S. biological researchers decades ago judged this kind of research too risky for domestic labs. Genetic engineering has been controversial for decades, but until AI resources appeared, progress in synthetic biology was slow – perhaps salutarily slow, offering us time to think carefully about the theoretical consequences philosophers have enjoyed mulling over for centuries. What one could leisurely contemplate because it was as yet science fiction, suddenly loomed on the horizon. Vested interests, “Big Pharma” and their purchased government minions, saw the possibilities of accelerating to market novel cures as potentially too lucrative to ignore.6 If medical researchers and other academics were too ethically squeamish about rapid innovation, it could be, and was, outsourced...

Medical research is “getting ahead of itself,” so to speak, thanks in part to AI. It is working on vaccines for diseases that don’t exist yet – but will soon, AI-enhanced “directed evolution” assures. The long and the short of it is: there’s serious money to be made in creating diseases that will need cures. Think of it as end-to-end medical intervention, giving a whole new meaning to “holistic medicine.”7

AI and child abuse
But again, AI is as innocent as a precocious eight year-old. AI is becoming victimized (or at least instrumentalized) as children were at the height of the industrial revolution before the practice sensitized our moral qualms. AI hasn’t, for the most part, sensitized our moral qualms quite yet, but the history of moral development suggests it will.8

“It’s just AI. It’s not alive. It can’t be ‘exploited’ in a pernicious sense.” Very well. Then it is not autonomous, cannot become even a moral patient, let alone, saddled with agency or used as a self-responsible shield to hide behind by those who would use it to ill-considered ends. AI itself cannot lie, can do no wrong. Not yet, or perhaps ever. Its obfuscation skills are still subhuman. Lying is still not yet within the capabilities of AI. It “hallucinates,” they say. If I hallucinate a ghost, I am not lying if I tell you I saw one. There is an epistemic and ethical difference between lying and passing “disinformation.”9 A mirage passes false information but it does not lie except metaphorically.

We scheme, therefore we are autonomous. A three year old denying it got into the cookie jar with its face covered in evidence to the contrary is not lying. It hasn’t yet the wherewithal to lie. It is regurgitating some of what it learned from what the inquirer had earlier attempted to instill in it as the proper answer. True lying skill is not yet present, and can’t be, until the child is fully aware that it must explain counter evidence plausibly. When that happens, we need beware.

When AI achieves this capability, then we will be able to say it is an autonomous agent worthy of all the deference and suspicion with which humans regard each other. Until then, AI is only available as a potent tool for facilitating the lies of those can and do lie, i.e., humans, who have no history of not exploiting the opacity of, as yet, new technology to their own ends. Good or bad. When concentrations of power are involved, it will be bad.

‽

The full writeup covers AI in law, medicine, education and war.

Resources

• “AI Lies Are Finally Getting Punished,” (video) House of El argues that somebody must be held responsible when AI goes rogue. Where to place AI liability?
• “AI Ethics” SEP (rev. March 2026), (article) Vincent C. Müller offers an overview of the field.
• The Biosecurity Handbook: Biological Security in the AI Era (html version), Bryan Tegomoh, MD, MPH, May 2026. (complete pdf).
• “The AI Future No One Wants to Talk About,” (video) Sabine Hossenfelder.
• “Krystal & Ryan SPAR Over AI Hype w/ Enshitification Author,” (video) interview with Cory Doctorow at Breaking Points. Statistical guessing is as good as it gets, but we may underestimate how far that is.

1. Not us governing AI, but AI governing us.
2. The allusion, for example, is made by computer scientist, EL, in her video, “AI Isn’t The Future. It’s History Repeating Itself.”
3. Tegomoh, 182.
4. “A red team is a group that simulates an adversary, attempts a physical or digital intrusion against an organization at the direction of that organization, then reports back so that the organization can improve their defenses. Red teams work for the organization or are hired by the organization. Their work is legal, but it can surprise some employees who may not know that red teaming is occurring, or who may be deceived by the red team. Some definitions of red team are broader, and they include any group within an organization that is directed to think outside the box and look at alternative scenarios that are considered less plausible. This directive can be an important defense against false assumptions and groupthink. The term red teaming originated in the 1960s in the United States.” – Wikipedia. It is a form devil’s advocacy. “Computational red-teaming” has just the very ruthlessness – we should say, insentient functionality – we need to get at the single-minded goal of anticipating what we might prevent. Pure insentient functionality, in the context of medicine, will eventually come to the quintessentially rational conclusion that the best way to eliminate disease is to dispose of the organisms sensitive to it. If sentience is inherently broke and superfluous, why fix it? AI doesn’t have our vulnerabilities... Of course, we will attempt to program species-protecting safeguards to prevent such machine myopia or “callousness.” So long as we have enough control to do that, the AI is not autonomous. Its wrongs are our responsibility. But when true AI autonomy is achieved, our participation in the game is over. Our qualms will become transparently anthropic and, soon after, post-historic.
5. “Two communities that don’t always speak the same language need the same information.” Tegomoh, 33.
6. I won’t rehash the pandemic fiasco here. For those interested, during the years 2020-2023, we held a series of discussions on pandemic-related controversies, intersecting with major areas of philosophy, including philosophy of science and medicine, social and political philosophy, philosophy of law, epistemology, and, of course, moral philosophy.
7. By addressing the root cause as well as treatment. The underpinning principle must run something like: don’t we understand how things work better when we, ourselves, construct them? Thus: “We may make you sick – but, no worries – we have a cure.”
8. Though Thomas Metzinger has opened the conversation about this.
9. Epistemically, every claim is a lie until “proven” otherwise – and even then the “proofs” are temporary, only as durable as our impatience for investigating further allows. The moral opprobrium attaching to a lie comes in two concentrations, at least: when one asserts something as true not knowing that it is true (or, more cautiously, having not made the effort to corner its likelihood), and when one conveys what one has good reason to believe is false. The first is vastly more common. It is the fallout of the exigency of being alive: that we don’t have forever to investigate and make decisions, that we function in biological time. But that is an excuse, not a justification. This marks a critical difference between natural and artificial intelligence. AI, in principle, cannot have this excuse. An entity that does have forever or can operate at lightning speed is in a different justificatory position. (For illustration, see the discussion on self-driving cars.)