Schneier on Security

Clive Robinson • March 4, 2023 6:49 AM

@ MarkH, lurker, modem phonemes, ALL,

Re : Computational linguistics

“It’s organized around computational linguist Emily M. Bender, who has done her best to sound the alarm concerning harmful implications of this technology, co-writing in 2021”

It’s something I’ve been thinking about since the 1990’s and you will find the occasional example of my thoights on it in this blog.

My thinking was on the fact english was becoming the language of science and what harm that was doing to science.

The way people think is known to be intimately tied to the language they speak or more correctly reason in. Thus you would expect two people one thinking in German and one in French but otherwise of the same real world knowledge and intelligence to think differently on a theoretical problem.

But there are two questions that arise the first is,

“What if it’s a practical problem?”

Would you expect to see any real differences? Probably not depending on the complexity of the practical problem.

The second question is,

“What differences would you exprct if they both thought in english?”

It’s actually in part a diversity and culture question but the roots grow so deep they effect even hyper-rational thinking thus activities.

But to move on I also in the 1990’s got into the fundemental asspect of this argument with “user interfaces” that is what model to follow the “learned” or “expected”.

Back in the 90’s the worlds largest machine, spanning the globe and reaching into geo-stationary space and back was the “telephone network” it was a time when DTMF / key dialling had more or less replaced Pulse / Rotary dialing.

With few exceptions the human interface was the handset, and hook / base set that had a nine or twelve key keypad and one or two buttons that flashed the line or did redial.

To use the phone you “learned” the interface like the use of the 9 key with a PABX to get an outside line and various uses of the flash key. The only feedback from the phone interface was the audio tones on the line, that you had to also learn. The interface to a car was also “learned”.

The other interface is the “expected” interface that back in the 1990’s was,

“Ask a human who had ‘learned’ the interface to a machine.”

Some called it the “managment interface” the incorrect assumption being a “master-slave” relationship.

So a manager might ask their assistant to place a call for them, rather than lookup the number etc. Or a member of the public might ask a “gate-keeper of service” or “receptionist” to call a cab for them or similar.

Likewise as a passenger in a cab you have an “expectation” of not just that the driver can opperate the vehical but also have not just local geographic knowledge, but also local temporal knowledge of traffic flows and paterns so that you do not get stuck in traffic etc.

The thing about the “expect” interface is you want not just the “learned” interface issue taken away but you also want “knowledge” to be applied to the “learned” interface for you.

But does that “knowledge” need “inteligence” to have and apply. The answer is an emphatic NO.

Back in the 1990’s the first usable “voice interfaces” came along running on the likes of high end IBM-PC boxes running SCO or similar Unix (Microsoft TAPI was still not in the “sharpened stick era” back then).

You could phone up such a box and say a persons name and the box would dial the extention for you. Thus it was an “expect” interface all be it low on knowledge and learned interface function. But it in no way had intelligence of any kind, it was statistical models all the way.

As far as “expect” interfaces have gone we got Siri and friends, that had a larger knowledge base and improved statistical models. But again no intelligence as could quickly be found out, sometimes expensively.

Modern ML arises from the 1980’s “Expert Systems”(ES) that although they had no intelligence could appear capable of “intelligent action”. And that was the problem people not of the domain gave the ES defference as though it was intelligent, because it could do what they could not.

Now we have LLM ML that is in reality no more than an ES with better statistics and a much larger knowledge base.

The difference is that the LLM ML systems have an “Inference Engine”(IE) that looks for “signals in noise” or more correctly “finds patterns in non sochastic data” it then puts those patterns into a feed back loop that acts like a “lossy integrator”.

Yes they make very fancy ES systems with faux human interfaces, but they are neither human or intelligent in any way.

As for the “singularity” that young technologists and hippy dippy old AI guys think will happen…

A little lesson in humans and turtles/fleas and Red Queens is required.

The US,expression “turtles all the way down” and the English equivalent “lesser fleas” are about infinite regression. In philosophy you find nearly all the arguments are of this form untill you arive at the “homunculus” or “little human” which is the “axiom” you have to swallow to stop the infinite ride.

The ride by the way goes both ways, only in the upwards direction the homunculus is imbued with notions of “betterment” that is the human Id needs something better to strive towards. The important point that nearly every one misses, is that we need that homunculus to be almost within reach, which means it has to improve as we improve, so we always chase what is almost but not quite in reach.

The Victorian author Lewis Carol, no mean logician and philosopher of his time, wrote about it in his book “Alice through the looking glass”. Only he chose to set it against “religion” where the “Red Queen” was actually a tyrant, and the race was designed such that no matter how hard you ran, you would never move forward (ie the evil in the “Protestant Work Ethic”).

The “singularity” is the “techno-deity” it’s whole point is to entrap people into a belif system that is compleatly false. Like the religious rapture, it’s not going to happen, it is a method of extracting wealth or it’s equivalent work, to be used for power and ultimately control.

The danger is it has a flip side. A “learned” interface needs the user to improve and develop themselves in knowledge and skills, in so doing they improve their independence on others and their intelligence they move forwards. An “expect” interface however for the majority does the opposit, it discorages development of knowledge and skills, and increases dependence. Thus “the machine” will control them, but that leaves the question of “who controls the machine” that is where is the true “Directing Mind”(DM) as it’s most certainly not in those statistical models nor can it be as they are from “things done” thus in the past.

So the reality of the “singularity” if it happens, is not that technology will become intelligent or better than us. No it’s because we will due to lazyness sink below our own past achievements. The fact we are clearly sleep walking into this guilded cage should be a clarion wake up call. But it’s not.

To many believe that having the worlds knowledge at their fingertips makes them some how better. It does not, firstly you can never read it all, nor can you index it in a way you can find usefull answers so it is locked away from you by your own inabilities. Trying to jump over those inabilities by asking questions of a machine will not in any way improve your critical skills to evaluate. Worse the machine will tell you what it has been told to tell you, either by individuals seeking control or worse by the ignorance of the masses asking the same or similar question thus making it a signal much above the others…

Clive Robinson • March 4, 2023 1:53 PM

@ lurker, ALL,

Re : New words.

“Some new characters, many made up of pairs of seemingly unrelated characters.”

This is not just Chinese it happens in.

The problem is do you translate,

1, The sound of the word
2, The meaning of the word
3, Create a new lable to the meabing of the word.

I know because my name is not easy for many oriental speakers to say that the sound of some words will not translate.

To a lesser extent some meanings are absent so can not translate “Hydrolic Ram” causing some issues.

But the creation of a new lable is most often for “slangs” and “jargons” and other in-group language deliberately designed to make out-group individuals obvious or keep things hidden. Most often you get to hear about “criminals v. Police” where the code the criminals speak is unknown to out-group individuals thus evesdropping is at best unreliable.

One example being the so called Rhyming-slang where you say a a non rhyming word of a couplet that does rhyme. So you would hear “a richard” and unless you knew the couplet was, “richard the third” you would not be able to guess. However the number of words that rhyme is normally more than one… In this case “a richard” is “a bird” not what many assume. Thus “up the apples” via “apples and pears” becomes “up the stairs”.

The thing many forget is words are labels to meanings and used to reduce communications bandwidth, as well as improve reliability of communications in noisy environments. Thus a particular meaning can have many labels.

There is a debate bubbling away about “slur words” that is words that are proxies for insulting meaning. For some reason some believe –I personally think quite incorrectly– some how changing the label makes the slur the less.

It’s why we see “they said the X-Word” where X is one of several letters often the initial letter of the word deemed to be the slur word. The fact we know what the original word was and in all probability the actual slur so why is the X-Word aby the less a slur.

History shows us the danger of this, there is a word which can mean the outfall of the digested process. Because it was baned in 1950’s ScFi and the like the word “sugar” was substituted and contrext usually revealed it to be a “cuss word”. Now however people actually say “sugar” ad a cuss word so when hearing “Oh sugar” we kind of know it’s not an exhortation to the goddess of sweetness…

So how long befor the X-words just become a new considered just as bad pointer to the slur meaning and importabtly used as such?

Clive Robinson • March 4, 2023 6:09 PM

@ vas pup, ALL,

Re : Orwellian New-speak.

“That is why I am strongly against any mandatory political correctness because that is reminds me Inquisition activity in Middle Ages when reasons start to substitute by emotion – just personal opinion.”

You are right to be concerned, and you are not the only person by a very long way to think a bad idea.

But it’s not just the Middle Ages, the same has happened more recently and in many ways is deeply documented as well as deeply engrained in many peoples psyche. Worse it is become a tool for those seen as being not just on the fringe right of violent politics, but every day mainstream consetvative politics as they try to over come their increasingly disliked behaviours.

This modern history starts with the work of author and social commentator George Orwell. Because of his very poor health, he lived at the bottom of Pond Street Hampsted and ended up working for the “British Broadcasting Corporation”(BBC) during WWII. Whilst there he witnessed the various forms of propaganda both directly and by the way they were linked together by amongst others ex Daily Express journalist Sefton Delmer who had amoungst other things interviewed Adolf Hitler prior to hostilities.

Sefton put together plans for what is now called “black propaganda” some of which involved the most powerfull transmitter of the time, known as Aspidistra down near Tunbridge Wells in Kent and later used by the BBC and UK “Diplomatic Wireless Service”(DWS) and mentioned a couple of times on this blog.

For Black propaganda to work effectively it has to be backed by both Grey and White propaganda. The BBC put out only White propaganda and became a trusted source of news around the world. However what White propaganda was put out and when was coordinated to make the Black Propaganda look not just believable but highly creditable.

The trick, tell the truth with a twist, change meanings of words and slip in unchecheckable innuendo and similar as questions and similar (what we might think of as click-bait titles to factual information twisted to give a very different perspective or faux-news on the “One man’s meat is another man’s poison” principle).

You will find much of what George Orwell witnessed in the book he wrote in 1948 that got titled 1984. The scary thing, is he wrote it as a warning to society about what was coming… However from the 1990’s onwards UK and other politicians have used it effectively as a training manual.

Even more scary is how nearly every term Orwell coined such as “Room 101”, “spin”, etc has become not just everyday parlance but titles for TV entertainmebt programes like “Big Brother”.

Such words and phrases have had an immense effect on the citicens in nearly all First World and Western Nations. The original intent effectively lost.

Clive Robinson • March 5, 2023 12:43 PM

@ Bruce, ALL,

Banning TicTok App won’t help

https://gizmodo.com/tiktok-ban-joe-biden-28000-apps-sdk-data-china-1850174019

“We Found 28,000 Apps Sending TikTok Data.

TikTok’s software development kits could undermine Joe Biden’s order to stop internet traffic flowing from federal employees’ phones to TikTok within 30 days.”

Not exactly surprising…

Trying to ban the apps individually won’t work as nobody has the time to go through a list that long.

Asking the “Walled Garden” owners to remove the apps won’t work either.

The only thing that will work in theory is for every phone to be recalled and then wiped / factory reset, then all phones regulated / forced to use Approved Apps only, from a US Gov run “Walled Garden”.

That would mean the phone manufacturers giving up control and their lucrative control… Not going to be an easy sell…

But it would still not work…

Because TicTok has a web pressence as well just like all the big Silicon Valley Mega Corps of Social Media. And they collect all the same data if not more than TicTok does…

So maybe blocking the Internet to China ? No that won’t work either.

Because those mega corps The Facebook, Twitter, LinkedIn, and Alphabet not only collect way more than TicTok, they allso sell it all to data brokers who then sell to who ever can pay the price asked as there is no law stoping them and several laws supporting them…

That is it is US legislation that alowed such Meg Corps to take anything they could choke out of people and comodatize it and sell it to who ever had the cash. So US “Free Market” behaviour in a Capatilist compliant system with no consideration of harm…

So even eradicating everything TicTok off of the face of the planet is not going to stop the problem of US Citizen data flowing to China. It was happening before TicTok, and it will continue after TicTok via US Corps and US legislation.

Clive Robinson • March 5, 2023 8:26 PM

@ Winter,

“There is currently no reason to think our tree dwelling ancestors used a “language” different from that of modern apes.”

You will note, that I am quite skeptical of those arguments to put it politely, but others insist on making them.

As for,

“People’s language reflects their thinking, and their language will change with their thinking.”

Err no, you are thinking of jargon, and that arises for reasons of secrecy or in-grouping.

The best you can actually say about a language in common use is,

“People’s language reflects that of their ancestors and the environment they were in”.

If you believe what some linguists think then two events that radically changed the average number of words we use today in the english language are.

1, William Shakespeare “wordsmith and playwright”.
2, Victorian Gentlemen who avoided religious strictures of the Universities and started natural philosophy (physics) “science” and the birth of engineering.

In the later case it was the change of environment from agrarian to industrial that forced the need for new words as new meanings had to be communicated. However there were others such as “phlegmatic”[1] that turned up as a “status gap” measure of gentility in the middle classes of those who lived on unearned income.

[1] The word “Phlegmatic”,

1, Resembling, consisting of, or producing the humor phlegm.
2, Having or showing a slow and stolid temperament of someone made apathetic or sluggish from significant phlegm.

Whilst the modern form of the word became popularised in the Victorian era, we know it goes back further and is found used in physiology atleast as far back as the mid 1600’s and probably further (into Shakespeare’s time). Where it was the pallor and cold, moist, humor of the body. Given that it is derived from the Greek “phlegma”, for “flame” (as in burning), it probably was originally a literal reflection of the fevers and inflammation followed by chills that respiritory diseases such as colds, flus, and worse all to often bring.

Remember this word when you next get blocked sinuses, you will unfortunately come to understand it, especially as it will be easier to say in that state…

Clive Robinson • March 7, 2023 12:37 PM

@ Warren, MarkH, ALL,

Re : Perfectly Secure v. Secrecy.

“Yet another “perfect security” claim in the wild”

The team is from Oxford Uni, which is a leading research organisation in the UK…

But we already know how to do what they claim. It is quite possible, just harder work than people would like it to be. I’ve already described a system that will do it on this blog before to demonstrate the point that no regulation can stop “End to End Encryption”(E2EE) being used or even reliably detected by an observant 3rd party, and with care is also secure from betrayal of the first party (Alice) by the second party (Bob) to the third party (Eve).

It’s all about statistics.

For the encryption side you use the “One Time Pad”(OTP) or equivalent Shannon Perfect Secrecy[1]). Which gives a message with what appears to be random nearly flat statistics.

For the Stenography side you first analyse the statistics of the “unique one use” carrier (image, document, what ever using the less well known Kullback-Leibler divergence[2]).

There are various ways to do this many of which are already done for you in compression algorithms.

Knowing the statiatics of the carrier you build what appears to be a second encryption stage, which modifies the effectively flat ciphertext statistics from the OTP output to that of the carrier.

You then insert the statistically modified OTP output into the carrier and Alice sends it to Bob. If Eve captures the communications if she analyses the image the statistics are as expected (remember as the image is unique Eve can not compare the sent carrier modified with ciphertext in it to the unmodified carrier)

The problem for Bob is two fold but not exactly difficult. Firstly he has to analyse the carrier and determin it’s statistics and build the appropriate decoding matrix. He then pulls the encoded ciphertext from the carrier and applies the matrix to get the OTP cipher text back. Having done that Bob recovers the plaintext by applying the OTP in reverse.

So now you can see it can be done, you can now read the paper knowing that it is possible.

I in the meantime as I’m going out tonight with a friend for dinner, I will read the paper tommorow when I’ve a few minutes spare to see what their basic system is and how it works.

I’m not expecting any surprises either way.

[1] https://en.wikipedia.org/wiki/Semantic_security

[2] https://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_divergence

Clive Robinson • March 7, 2023 8:03 PM

@ MarkH,

Just in from my evening out and I see your,

“the authors’ point of reference is the aggregate distribution of the universe of possible covertexts from which one was selected.”

If you reword it that is the “all messages are equiprobable” proof of security of the one time pad.

If you look back at my comments back when talking about proving E2EE could not be stopped as plaintext could be selected by the use of a OTP output by using standard plaintext such as the opening salutation {Hi, Hello, Howdy, goodday} qives you two bits, and “We should meet up for a XXX” where XXX is one of {drink, tea, coffee, beer} gives you another two bits and so on[1]

Is what you are describing.

I guess at this rate the paper is going to make a dull read as you make it sound like they are trailing in my footsteps by quite a ways…

[1] Even if the adversary Eve suspects this is the system, they can not prove it because of the use of the OTP robs then of corelation. But even if Eve knows the system because Bob betrayed Alice Eve can still not prove it because again the use of the OTP breaks any correlation. As long as Alice takes care not to corelate any of her actions with the actual plaintext the Eve still has nothing. Because Alice can just say Bob is mentally unwell or at a push provid a constructed pad that gives an entirely different message. Hence the real message is fully deniable. To make it worse a compression system via “code book” makes deniability even easier, especially if the only copy Bob has is in his own hand writing.

Clive Robinson • March 8, 2023 3:06 AM

@ MarkH, ALL

Re : The Oxford paper.

“I saw nothing in the paper about an adversary’s ability to recover plaintext. It focuses exclusively on an adversary’s ability to discriminate stegotext.”

It’s quite a bit more subtle than that.

Look at the right hand side of Fig 1, you see the “Receiver”(Bob) and the “Adversary”(Eve) both recieve the stegotext(S), but Bob also receives the Private Key(K) which Eve does not.

Now look at the definition for K it is exactly the same as for a “binary OTP”.

Likewise the definition of the stegotext S is the same as that for the “OTP ciphertext”.

They are the same function and have the same logical consequence, all that changes is the “labels” they have chose to call K and S, and we could as equally call OTP KeyMat and Ciphertext.

Thus the lables “plaintext” and “stegotext” are likewise equivalent. So consequently,

“they wrote perfect security, not perfect secrecy”

It also makes “perfect security” and “perfect secrecy” the same when viewed as a process to an end (recovering plaintext from ciphertext or demonstrating is plaintext is covertext).

But the thing about an OTP that every one forgets is that,

“All messages are equiprobable”

Has a flip side, which is what I point out with the “OTP deniability”

“Logicaly every ciphertext from an OTP even if it reads as plaintext is infact any and all messages of the same length as the cipher text.”

The paragraph above this has 112 visable and 25 white space charecters for a total of 137. It contains every message in the universe that is 137 charecters or less long including what it reads as. Because it’s the private key K that decides what the message realy means. So if K is all zeros then the message is what you read as the plaintext. But depending on K it also is by definition a saucy message to some one you fancy, even though you have not thought it up yet…

So by definition every piece of plaintext is also a covertext for every “stego” message…

Also some subset of all those covertexts C look like plaintext under any statistical test you chose, that is they will have the same distribution within your ability to measure it.

All they authors of the paper have proved in this respect is that “all stego messages in the set M are equiprobable, including a null or all zeros stegotext that is genuine plaintext, and stated that some will have the same statistical properties as others. None of which is in doubt, or ever has been since Claude Shannon effectively pointed it out.

Thus all they are saying is that there is a subset of covertexts that as far as your measurments are concerned have exactly the same statistical properties as messages you think are plaintext not covertexts thus you can not tell them appart by measurment, tgus cannot prove which they are.

I’ve mentioned this property in the past several times when talking about E2EE[2], and stated you can send secret messages that look like plaintext and the advaserial third party Eve has no way to tell let alone prove.

If you look at my posting above that has a couple of sets {} defined. It should be abundantly clear that it is a very simple system to send bits of information covertly where the statistics are indistinguishable from plaintext because they are plaintext. It is not the first time I’ve posted this scheme, thus I’ve already shown by the equivalent of a “grapical proof” that you can do this, and unkike this papers authors I did it simply so it would be easy for anyone with no more than a K12 school education to grasp…

So in this respect they have not proved anything new, as I have already done so ages ago and posted it to this blog for people to think about. And for all you, or I, or any other reader of this blog knows, one or more of those authors have seen it and effectively copied the idea consciously or otherwise (it’s been shown to have happened with other researchers who were once part of Cambridge Computer Labs, for which they still owe me and @Bruce a drink[3]).

But unlike them I don’t claim it’s original to me, I’ve never bothered to check. What I have done is said in the past my thinking was a logical consequence of following through on Claude Shannon’s work, and the BBC transmissions of “now some messages for our friends” to SOE officers and other agents in occupied Europe during WWII.

Oh look up who the K and L are of that test[1] are and who they were working for “part time” when they made it public in the 1950’s… I actually have a signed copy of Solomon Kullback’s book “Information Theory and Statistics” in my dead tree cave, that I was given as a present.

[1] For those who want to know more about the “Kullback-Leibler Divergence” but can not face all the domain specific terms. As long as you understand Pythagoras’ therom and can code in Python, then this will prove of interest.

https://bigdatascientistblog.wordpress.com/2017/09/11/a-simple-introduction-to-kullback-leibler-divergence-through-python-code/

[2] I’ve also mentioned it in other places than this blog, most recently in,

https://www.lightbluetouchpaper.org/2022/10/13/chatcontrol-or-child-protection/#comment-2425357

So there is more than one place the authors could have seen it.

[3] I have repeatedly said on tgis blog in the past, people can use my ideas if they wish, but with two conditions,

1, They acknowledge both our host @Bruce and myself.
2, And if they meet either of us they should buy which ever of us it is two drinks. So that which ever of us it is can by the other a drink or more if we ever meet in person.

(yes someone did mention that we should both join “buy me a coffee” or equivalent.)

Clive Robinson • March 8, 2023 3:48 AM

@ Winter, JonKnowsNothing, ALL,

Ahh “Zombies” every shed should have one.

(see https://en.m.wikipedia.org/wiki/Shaun_of_the_Dead if you don’t get the reference).

Clive Robinson • March 9, 2023 7:04 AM

@ JonKnowsNothing, SpaceLifeForm,

Re : More on LC19 and the security cost to society.

Have a look at an article about a new report on C19 after effects,

https://arstechnica.com/science/2023/03/having-covid-ups-risk-of-many-long-term-gut-problems-study-finds/

It appears that C19 may be bringing IBS and worse –including pancreatitis that can kill you– to people that have had C19.

Something as the article notes,

“The study likely confirms what many long COVID patients already know all too well.”

As others know LC19 is now in reality a chronic disability condition with no known treatments other than a limited range of palative symptom reducers.

What makes the study interesting is it’s depth and breadth of sufferers and a very large effective cohort of control groups,

“But the analysis is among the largest and most comprehensive to evaluate the boost in relative and absolute risks, drawing on medical records from more than 11,652,484 people in the Department of Veterans Affairs databases.”

Whilst the medical research community are still pondering the whys, it should be noted that intestinal tract issues are not exactly unexpected with C19.

Because whilst in humans C19 preferentialky goes for the respiratory system it also shows up in the gut. In animals the related viruses are almost all gastrointestinal not respiritory and is in part why the animals survive, and get repeated infections, and form disease reservoirs so the pathogen is endemic.

I suspect we’ve a lot more “comming down this particular pike”…

But there is the question of societal security. In the US this is an especially vexed question most visable as “disability wealfare” but actually effects society in it’s totality.

Obviously working with many of these gastrointestinal disorders is going to be both difficult and require considerable consideration by employers, to the point employment will not be possible for many.

As some know, in the US if you don’t work your level of health care options are at best vanishingly small, and non existant for chronic conditions.

I suspect there is going to be very considerable tension over LC19 in the near future. Certainly in both the US and UK people can easily point the finger at certain individuals who are very largely to blaim for C19 mishandeling and what happened because of it. The fqct they and their supporters are in full and sometimes violent denial is a considerable societal security risk just on it’s own.

But societal security wise it gets worse, a lot lot worse and the ebtirety of US society is not just at rik but is provably harmed which will cause many other security problems. As long term readers know, I was raising the issue of pathogens knowing no distinction between rich and poor or geographical divide here long before C19. What C19 should have made clear even to the most ignorant, Health Care effects not individuals but the entirity of society with no doubt. If you alow a socio econmic group to be vulnerable, then they will get ill. Not only do they become a disease nexus, they also become the origin of entire community spread, which will hit rich and poor alike with no favours shown. Some US politicians died of C19 despite having arguably the best healthcare available to any US citizen as part of their “job package” thus this should have been a major wakeup call that there is something deaply wrong with the way healthcare in the US works. Effectively if it’s not universally available then society at all levels gets sick and dies before their time. Which might explain in part why the average life expectency in the US is dropping, but going up in other Western First World economies, where healthcare is a fraction of the cost of the US system and universaly available.

Clive Robinson • March 9, 2023 9:52 AM

@ FA,

Oh one other thing, if you look at what @MarkH says in his part 4 you will find,

“The authors claim about 0.9 bits of payload per bit of message entropy, or roughly one bit per character of English text”

Do you see why that actually makes the Oxford system much much more susceptible to deliberately induced “channel noise” than the system I described?

Read the rest of @MarkH’s part 4 and you will see he is aware that the KL test is alone unsuitable, or the resulting system would be to fragile.

You realy should atleast try comparing and contrasting when making the statments you did.

Things you realy should have picked up on and mentioned but strangely you did not…

Care to explain?

Clive Robinson • March 9, 2023 1:35 PM

@ JonKnowsNothing, SpaceLifeForm,

Thanks for the update on BoM&D and the proliferarion of “Swedish techniques”.

What scares me is in the UK there are yet more new NHS policies from central government that are so ill thought out they should be considered professional misconduct if Drs implement them.

I can name atlrast one practice where the “call back” for emergancy medical issues is “four weeks” but won’t happen untill six or more… Each missed deadline is excused by “your Dr was unwell” or similar. On investigation it turns out “the practice” has laid off the fully employed Drs and replaced them with “temp type locums” who get paid so little some find it better to do “nurse style” bank work which kind of shocked me because there used to be a strong demarcation between being a Dr and a Nurse with different proffessional bodies etc.

But the thing is “your Dr” is a compleate misnomer, you are deliberately discoraged by UK Central Government policy from forming a “relationship” with a Dr.

Which I find worrying, because a Dr you have a relationship with knows you well enough to either diagnose what is likely wrong with you, or how severly it is effecting you before you get to sit down. Something that does not happen with voice / video calls.

Then there is the “you’ve committed a serious criminal offence for which there is no defence” issue. Drs have been told to use a “Photo sharing” system run by a third party service provider where you upload a picture of your effected part… Well what ever you do say NO if it’s someone under 18years old because it can be considered CSAM under law that has both mandatory reporting and a no defence legislation stricture. So photographing say a nasty graze or similar on your childs hip or bottom from comming off of their push bike or park play-frame etc will get you arrested, convicted and forced to sign the sex offender register… Oh and that still applies for say a teenager effectively taking a selfie of their afflicted body part, even though they are technically not “legally responsible”.

It’s a central government ministerial madness that is getting worse not better on an almost relentless path to full on lunacy.

Clive Robinson • March 9, 2023 3:23 PM

@ MarkH, ALL,

Re : Photos snatched and not returned.

“… there was a very disturbing example in the U.S. which came to light a few months ago.”

I’d sort of heard about it but the source was not always shall we say trustworthy due to excess editorial leaning politically.

At the time I could not find other refrences so I treated it cautiously.

I’m glad he was not prosecuted, but it does show how easily you can become a criminal in some organisations eyes, even though eventually your name gets cleared your life does not return to where you were…

The other point to note is “backups” and similar like YouTube videos. If you do not have the original uploads 100% under your control there is an increasing probability they will be snatched away from you and you will have no redress.

Why increasing probability? Because the folks on The Hill want to chew more than Zuckerburgs ear, they want fresh meat, new legislation and masive fines to crow about.

So shareholders are going to insist that social media and similar will avoid scandle to avoid attracting the eye of rapacious prosecutors on the make.

As we’ve seen with Twitter and Moderators politicos can get insistant and I suspect the EU will get it’s way and Hell-on Rusk will have to throw even more out the pram.

[1] From what some pundits have been saying not only is Twitter sinking like a brick in the ocean, Hell-on is having to burn currency just to keep it in sight. Thus stories are saying it is worth less than nothing and never will be anything else. I suspect they only mean in the short term, but if I had money in any Social Media or related Organisation including Microsoft and Apple, I’d be looking for a safe haven to wait out the comming storm.

Clive Robinson • March 10, 2023 3:48 AM

@ Bruce, ALL,

New twist on APT being persistant

This is one of these attacks that are,

“Obvious with hindsight, but not obvious if you have not thought about it with foresight.”

The APT code survives firmware updates because the developers of the target device were not thinking[1] with foresight when they developed it.

It effects some SonicWall products and is punted as having a Chinese nexus by Mandient (make of that what you will it’s not important technically).

What the APT code does when the device becomes infected is instal a loop that every few seconds checks if a firmware update has started on the device.

If it has the APT code waits untill the zip file has finished downloading.

If it has it unzips it adds in it’s APT code, and zips it back up.

Then alows the update process to finish with the now infected zip file.

Yup with hindsight it sounds obvious, after all it’s a varient on Ken Thompson’s 1984 Turing Award speech that outlined the “Trusting Trust” attack[1]. But if you examine a lot of “Signed Code Update Processes” used in embedded systems you will find this “security hole”[2] in them.

As so many make this “security hole” mistake, they should consider atleast one way to solve it[3] to decide the “Why? and How?” of their closing it.

You can read the boring stuff 😉 about this specific APT attack at,

https://arstechnica.com/information-technology/2023/03/malware-infecting-widely-used-security-appliance-survives-firmware-updates/

[1] What few realise about countering “trusting trust” attacks is it’s easy to get wrong. In fact they nearly always become an imposible regression problem on current computer architectures. To see why think about the load and linker processes on a computer to put an executable in RAM. It’s incorrectly assumed to be a trusted process, so once the executable is loaded into RAM it’s assumed to be trusted from then on. So no further checking during execution is carried out. (the “loader-linker” is a Shannon Channel, and the actual code execution is the transaction).

As we all should know from certain types of malware, the code in RAM can be changed or overwritten at any time and in conventional computer architectures won’t be detected (hence memory tagging of CHERI). But as RowHammer proved beyond any doubt, conventional hardware and even tagged memory will not stop RAM being illicitly altered during execution (it’s one of several “bubbling up” attacks for which we have few or no defenses for currently).

Having known about the RAM bit flip issues with radiation and how Parity based ECCs were not reliable back in the 80’s when involved with the design of Space Payloads, it came into my thinking and in part gave rise to “Castles -v- Prisons” and the notion of “Probablistic Security”.

[2] I’ve said it before quite a few times but people don’t pick up on it. Code signing especially as many implement it has major problems,

“It does not do anything more than sign an archive of the code at a point in time.”

And also a lament I’ve made since the mid 1990’s about several security functions (but especially about online banking),

“It Authenticates the channel, not the transaction.”

In the case of “Code Signing” it opens a significant security hole because it authenticates,

“A hash of the code archive not the code it’s self.”

So once past the “authenticate the hash” stage you’ve authenticated the delivery channel not the payload.

Think of it in a physical security analog as,

2.1, Checking the delivery code for a box at the loading dock door (but not the contents of the box).
2.2, The box is then put unopened in the building internal mail system.
2.3, Intetnal mail open the box to check for bomb/chemical attack or illicit goods (but do not check the BOM on the delivery note).
2.4, The final recipient gets the opened box thus potentially tampered with contents some time after when it’s put on their desk.
2.5, Unless the final recipient checks the contents against the original order, they won’t know if what is on their desk is what they ordered.

Which is why some of us do check the contents of the box against the original order at the door infront of the courier or sign it “unchecked” if they won’t wait (Amazon for instance are known to put the wrong stuff in a box such as a cheap disposable film camera, not a very expensive Digital SLR).

But many code signing systems designers only do the equivalent of step 2.1, so any old piece of code can subsequently overwrite the original signed hash payload. And the later stages of the update ie the actual installer process will not know or care as long as the code file is in the right format for the actual installer process (in this case a zipped file).

[3] To understand and design the solution should be obvious once the process failure[2] is known and understood. But it’s obviously not to most code signing system developers, so we can assume either the process failure is unknown to, or depreciated by the developers. So to go over it,

The hash check is simply a Shannon Channel process, through which a payload of the code is delivered. This payload is then pased to the transaction which is another independent process which is or is part of the actuall final code install process. As these processes are,

“Sequential, not Atomic processes”

If an attacker can get to the payload after the archive hash check, but before the final install process then it’s “game over” for security. The solution is to make the two sequential processes of “code check and final instal” a single atomic process, with a “brick proof” abort process.

That is the “signed hash of the archive” is the wrong security process to use in “code signing” you actually need to use a form of cryptographic chaining inside the code which is checked continuously in the actuall install process.

I designed such as system for “Broadcast Warehouse”(BW) for their first “Audio Processor” way more than a decade ago. It used a form of chained stream cipher over blocks with a strong MAC as the BW audio product development leader had four main desires,

3.1, Prevent “bricking” of the device.
3.2, Tie the download to an individual device.
3.3, Obfuscate the runtime code, to prevent analysis.
3.4, Keep the size of the secure update runtime code to a minimum.

The reason for 3.4 is not as critical these days, but back then it had to run on a low cost low electrical power microcontroler with a lot of other runtime system code, so resources were tight.

Hence step 3.2 which has the advantage of using a “shared secret, root of trust” so you don’t need to use Public Key signing and the massive ROM and CPU cycle penalties involved. Giving three advantages, firstly a much lower cost microcontroller could be used at a lot lower electrical power thus heat. So secondly it could be fully encapsulated within a security mesh system embedded in quartz dust loaded epoxy which is regarded as being nearly “tool proof”, effectively as a “mini HSM”. Thirdly it also gave “revenue and theft control” as getting updates for the device was a paid for service for each device, and devices reported stolen could be bricked if the actual original purchasor wished that it be done.

Clive Robinson • March 10, 2023 5:09 AM

@ Earl,

Re : Chinese and Secure Comms Sats.

“I think that Satellite based QKD is, at least least for the time being…”

The problem with that is two fold,

1, We only know what the Chinese tell us or we can surmise.
2, The development of Quantum Comms systems moves in fits and starts.

So whilst in the Open Community we can guess like we can with the NSA and GCHQ, we actually don’t know what progress the Chinese have made.

I opted for talking about entangled particles for two reasons,

1, It is as far as we know the most secure system.
2, It is relatively simple to understand.

So people can reason about it.

If you look back at the thread that has comment about the Chinese balloon flying over the US I did talk about the practical realities of getting data back to China from the balloon securely.

One of the problems about posting to this blog if you do it regularly (yes that is a hint 😉 is that there is a very wide range of ability in any subject area in the readers. So you have to pick examples that are simple enough for someone with a reasonable high school education to understand, through to sufficiently complex that domain experts at the upper end of the research end of things won’t turn away. As I want to encorage STEM I tend to err towards slightly above the high school “science project” level.

Look at it this way, if I can get just two new people interested and go on to entering the game, then I’m ahead on points 😉

It’s also why I say anyone can use my ideas but with two conditions,

1, They acknowledge not just me but @Bruce our host as well.
2, If they ever meet either of us they buy us two drinks.

Why two drinks, well that way when Bruce and I ever do meet up, we can buy the other the appropriate number of drinks 😉

As I drink mostly tea it should be fairly cheap not sure what Bruce drinks[1] (I used to know Terry Pratchett and his taste in drinks was good quality brandy and various cocktails in the bar after hours, and sometimes in hours, as he noted “book signing is thirsty work” 😉

[1] There are sites like “bymeacoffee.com” and Ankesh Kothari’s Buy Me a Beer WordPress plugin, maybe Bruce should set one up 😉

Clive Robinson • March 10, 2023 9:51 AM

@ Bruce, ALL,

Remove this door device before it wrings you out

You might find this of interest,

https://arstechnica.com/information-technology/2023/03/go-ahead-and-unplug-this-door-device-before-reading-youll-thank-us-later/

Longterm readers know I used to design electronic locks, enviromental, and other hospitality industry security and privacy related systems for a company called “uniQey” that used to be Opposite the BBC Rehersal Stages in Ealing West London.

Thus I’ve had occasion to think about both physical and informational security of doors and their multitude of functions, since then.

Which is why I keep an eye on the likes of the horrifying for privacy Amazon Ring etc.

Well what can I say about The Akuvox E11 “video door phone”?

I could start with “words fail me” but that would not be very informative.

ARS Tech’s take on this brain dead design video door phone on steroids is,

“Go ahead and unplug this door device before reading. You’ll thank us later.”

Before saying that it,

“is riddled with security holes.”

So where to go from there? Well, you could say “don’t buy it” or “take it back for a refund”, but that’s not describing the horror of it, or necessarily what you can do (see below about agents).

For starters it’s “Network Connected” which is never a good sign as anyone with a modicum of “Cheap IoT Devices” esspecially those that got used for “Distributed Denial of Service”(DDoS) system attacks will tell you.

Which is why the usual advice on IoT devices is “segregate, throttle and put behind a firewall”, but with this device that is clearly insufficient. Why? Because as the article indicates, there are 13 serious vulnerabilities that have been found that,

“include a missing authentication for critical functions, missing or improper authorization, hard-coded keys that are encrypted using accessible rather than cryptographically hashed keys, and the exposure of sensitive information to unauthorized users.”

Enough badness yet to make you feel twitchy in the presence of one of these devices?

Well apparently the Chinese manufacturer is not interested in discussing thus resolving the issues…

But because they are inexpensive, look cool and alow remote access to their functions, quite a few buildings come with them installed by the buildings managing agents…

And because the managing agents can access pretty much all the functionality remotely and much of the communications are in “plaintext” and poorly or not authenticated, then so can crooks and worse have the freedom to play…

Read the ARS article for even more badness, but first a “moment alone with it” with a hammer and screwdriver appears to be good advice.

On a score of “one to ten” I think there is probably a very large negative number heading it’s way…

This is not a “product review” more a “run for your life” warning of the sort given in Zombie movies.

Clive Robinson • March 10, 2023 6:36 PM

@ MarkH,

“I suppose that train must have crossed the new 10,000 foot threshold; I would have guessed it to be shorter”

It may well have been shorter when you saw it than it started out, remember part of the issue –still not addressed– is that for “speed of operations” NS order the train carriges not in “the safe order” but by “destination” so that only one decoupling is required for each destination along a longer route (now two with a “banker” if it’s coupled).

Whilst distributing the “power” will help in locomotion my concern is “braking” the ordering is still not in the “safe order” for that, and I suspect that in the near future NS will have another event that will show this to be the case…

But what realy scares me is that the traction and braking the engines provide will be kept in sync by “radio”.

Accidental jamming or drop out of radio links is rather more common than people realise, especially at more than around a mile distance using VHF or UHF (ask walkie-talky users in urban environments about this).

NS has made it clear that rediculously long and dangerous trains are what their share holders want as long as adverse publicity remains low… So it’s not hard to see what is likely going to happen to train lengths in the future.

The problem is if you have multiple traction/braking units and one or more looses communications,

“What should it do?”

There is actually no real “fail safe” in a simplex radio system and even duplex systems are not going to be “reliably” fail safe except on paper (it’s something I’ve found out the hard way with designing RTU systems for use on ships for linking upto buoys for loading and unloading petro-chemicals off-shore).

Clive Robinson • March 15, 2023 4:35 PM

@ modem phonemes, JG4, Winter,

Re : reliable unreliability.

Part 2,

The reason as we later found out he was actually thinking of abstract cipher channels.

But the point is the mechanical linkage between two wheels is a Shannon channel as are any causes of information to transfere from one point to another, and even down at the atomic level this notion of information channels holds true.

So any system with parts that transfer information use Shannon channels, even if it’s only down to chemical changes on the surface of a bar of metal.

Which is what @JG4 was saying with,

“… the chains of causality in systems engineering are the same in all physical processes.”

Because the “chains of causality” all convey information from one point to another point (and often in reverse as I’ve pointed out in the past when talking about Bi-directional transducers).

Clive Robinson • March 15, 2023 4:42 PM

@ modem phonemes, JG4, Winter,

Re : reliable unreliability.

Part 4a,

When talking about signals engineers talk about “the energy per bit of information”(Eb) in a “spectrum of noise”(No) as Eb/No and how you can integrate it over time.

Clive Robinson • March 15, 2023 4:49 PM

@ modem phonemes, JG4, Winter,

Re : reliable unreliability.

Part 4c,

If you want to know more you can start with,

http://s_s_s-mag.com/eb_n0.html

https://www.dsp_related.com/show_article/168.php

(underscores in URLs added to get past moderation)

Clive Robinson • March 16, 2023 11:27 AM

@ JG4, modem phonemes, Winter, ALL,

Re : Shannon and Moore Paper.

I would not bother trying IEEE, Sciece Direct or any other of those sinilar shysters. They will either send you in an endless tail spin, tell you you can get it from the authors or want to charge you large amounts of money.

Which is kind of typical for such MaxProfit journal stealers of IP.

So try instead,

http://www.mriedel.ece.umn.edu/wiki/images/3/30/Moore_Shannon_Reliable_Circuits_Using_Less_Reliable_Relays.pdf

I’m not sure it’s compleat, because after 18 pages and a section on Hammoch Neyworks, it ends prematurely with,

(to be continued)

Anyway the first half dozen pages should give you the drift without having to dig into all the hairy math. They give you,

1, Some of the historical background of earlier work by von Neumman and from Shannon channels (last para page 3).

2, The basics with math most can still get their head around.

The paper makes it clear that it is “theoretical” rather than “practical” relays that are being used.

However it does not realy go on to amplify the fact that relays are standing in for switches, and that switches are standing in for Shannon Channels, which cover any kind of information conveying system including the neurons in your head.

Clive Robinson • March 16, 2023 7:47 PM

@ MarkH, ALL,

Re : Fault tolerance.

“[1] To me, systems achieving higher reliability than their components are by definition fault tolerant systems.”

Nope that is not fault tolerance.

It’s potentially but not necessarily “high reliability” which if any faults that do arise can be quickly fixed potentially makes it “high availability”.

Fault tolerance is not realy about the components of the system but how the system behaves in an environment.

Remember all systems will fail eventually without external input, worse even with external input there is the probability that all parts will fail at the same time[1] but it’s important how they fail that takes you into the domain of “fail safe” systems.

Earlier I posted the “Roosters Egg” philosophical question, which is actually like the,

Q : An aircraft crashes exactly on the border between to countries. In which country do they bury the survivors.

Sometimes you have to realise that the question being asked is wrong, and not try to answer it but just say “You don’t bury survivors!” or the equivalent. But that still leaves a problem, that is the question arose for some reason and more often than not it’s important to know why (something people forget when they get excited about AI systems).

When you accept that it’s not practical to design a system that does not eventually fail, you have to ask “What do we do when it does fail?”

This leads on to the notion of “soft failure” amongst other things. Put simply another system seeing that the first system has failed tries to get the over all system to fail gracefully in a way where the operators have time to do things the system can not, or was not designed to do, or failing that fail with the minimum of harm.

If you look at the design of systems that are inherently unstable in operation and can enter a cascade or chain reaction failure the design philosophy is to try to fail in a manner that would cause the minimum of harm.

But can you actually do that? The answer is “you have to make assumptions”. In large vehicles such as ships and aircraft you have the possability that the operators become incapacitated. In civilian aircraft the usual practice is to adopt a straight, level flight path. Which is fine untill the fuel runs out not far past it’s desired destination. What you don’t get told is what plan B is since 9/11. Which is to shoot the plane down whilst over a sparsely populated or non populated area, regardless of if there are live passangers on board or not.

Similar with ships though in some cases it’s to put the ship in a circular course. This is based on the assumption that the vessel will eventually become “adrift” and can thus be “salvaged” rather than pile up on some costline spilling thousands or tens of thousands of tones of cargo like crude oil along the coast line.

The thing is any “predetermined action on failure” is only a “best guess” and very unlikely to be optimal though computers are trying to change that on ships. That is the auto-pilot system knows where the ship is and has information on wind and tide. In theory if a failure happens it can make the same or similar decisions a human could given the same set of information.

The problem is the computer is not aware, it relies on sensors, if the sensors fail, it can turn what would have been a nusance into a disaster and has done in the past[2].

[1] Some years ago on this blog I had to explain what the difference was between fail states and the probability of those states occuring. Take a jumbo jet it has four engines if we assume each engine has a binary working state then there are sixteen states. The desirable state is all engines working correctly, the undesirable is all engines not working. As a design you try to make the first probability as high as possible and the second as low as possible. However the seccond state can not be avoided therefore you have as an engineer to design in mitigation or contingency depending on your given domain jargon.

[2] An aircrafts air speed can be judged in a number of ways, but primarily it is via pitot tubes. As passive sensors they do not in any way effect the aircraft and it’s ability to fly. So provided the pilot is alert and aware then they can use other sensory information instead of the pitot tubes,

https://aviation-safety.net/wikibase/wiki.php?id=213551

Not so the auto-pilot, which can make wrong determinations and cause confusion, where it otherwise would not,

https://www.computerweekly.com/news/1280096931/Final-cockpit-moments-with-blocked-pitot-sensors

Clive Robinson • March 17, 2023 3:25 PM

@ modem phonemes, MarkH, Winter, ALL,

Re : Shannons information view of the world.

“It seems then one has to find the analogues of these components in physical processes or chains of causality.”

Yes and usually it’s not particularly difficult to spot the major information exchanges,

“Just follow the energy or matter”

Information has no physical presence in our physical world of matter / energy and forces.

The way we interact with information to,

1, Communicate it.
2, Store it.
3, Process it.

Is by impressing it on matter or modulating it onto energy and interacting with the energy or matter.

So where ever energy or matter move some level of information is transfered in a Shannon Channel.

So often the hard part as far as security is concerned, is not finding the channels, they are alas everywhere. But assessing their information bandwidth capability and what information can get into that channel both under normal and abnormal operation.

The prime example of this was back in the late 1980’s and through the 1990’s with “smart cards” and information leakage through the power usage spectrum on the power supply leads. Also as I demonstrated that low power VHF through microwave EM signals could pass through or be reflected off of a chip and be “crossmodulated” with the CPU signals, and carry them quite some distance.

(The opposit was also true, if you could get a sync signal, you could inject fault signals via a modulated EM carrier).

Clive Robinson • March 17, 2023 7:46 PM

@ JG4,

Re : Dioxins found no surprise there.

“Levels of carcinogenic chemical near Ohio derailment site far above safe limit”

Colour me unsurprised both of us expected Dioxins to be there, the only question as far as it was concerned was “just how much”.

And at way more than ten times the level cancer is expected to happen I guess it’s just a question of time, and how long the politicians can keep saying “Nothing here to see” as the proverbial three headed dog that is the hound of hades[1] starts snarling at your leg or worse…

But I’ll be honest with you and say as bad as Dioxin is, it’s the other nasties that concern me equally and nobody is talking about those…

Just so every one understands every time you burn even moderatly complex carbohydrates and similar organics at the low tempratures you would expect in either an open or barrel burn, you will always get carcinogens or equivalent chemicals that will interfere with living processes[2].

Similar is expected with solvents and similar chemicals especially the chloronated ones such as carbon tetrachloride, chloroform and other anesthetics, Halon combustion suppressants and refrigerants and dry cleaning and de greasing solvents. Most of which form real chemical nasties if they get hot enough[3].

[1] Cerberus of Greek Mythology is a large hound with three heads decided to police the entrance to hades,

https://en.wikipedia.org/wiki/Cerberus

I guess the question is when is some joker going to post a “Welcome to Hades, population 2000+50halves” or similar.

[2] Heck we know about cigarettes being carcinogenic, like wise cooking food that like wheat or other grains beyond light brown, so a piece of dark toast for breakfast might taste nice but is apparently nearly equivalent to smoking a pack of twenty cigarettes. Likewise any vegtable that gets “over cooked” like some fried or roast vegtables (but heck they can taste nice).

[3] Most people have forgotton about “Carbon Tetrachloride” and Chloroform. Both were used as anesthetics up untill the 1860’s when their downsides started becoming apparent, such as increased likelihood of cardiac arrest and the damage they caused to internal organs such as the liver and kidneys became clear. They also got used up until the 1980’s as a “dry cleaning” or “grease remover” for clothes and light machinery. As carbon Tet was found to be an effective fire extinquisher, the brass hand pump extinquishers were in common use in WWII (WAF’s ofte nicked a bit out of them to use to clean clothes/uniform such that most of the extiguishers quickly became useless and caused several reports to go to the highest levels. Even though it’s got quite a few disadvantages it is still used and is known as Halon-140. However between 900 and 1000C it decomposes into that cheary little chemical, NBC Warfare agent phosgene…

Clive Robinson • March 18, 2023 7:22 AM

@ MarkH, JG4,

Re : Toxins in East Palistine.

As I’ve indicated since the start, my interest is in “ALL the toxins” from the burn in part because they all have different ways into living systems.

The quote of “life time exposure” is always a bit iffy because the effects are not linear with actual time or victims age.

But simplistically the younger you are the significantly higher the risk.

Part of which brings us to,

“the notion of cancer risk from soil contamination must assume some transport pathway”

For those that have very young children the one thing you see is just how much they pick up and put in their mouths, either directly as objects, or indirectly as dirt on their hands.

It was a point I was making to @vas pup[1] just a few hours ago,

“The bacteria is Helicobacter pylori (H. pylori) and around half of us have it in the mucus lining of the stomach after we ate-dirt or worse when we were very young children. Curiously few go on to develop ulcers, and it appears there is a “susceptible type” where the bacteria gets out of control.”

Similarly with toxic chemicals that act almost as catalysts in living organisms disrupting DNA and RNA and other systems. Some will show no adverse signs whilst others will get horrible diseases and symptopms and die sometimes decades before their time. Why this disparity is so, is still after years of research largely unknown hence it gets lumped under the “Lucky Genes” meme.

[1] https://www.schneier.com/blog/archives/2023/03/friday-squid-blogging-new-species-of-vampire-squid-lives-3000-feet-below-sea-level.html/#comment-419510