Schneier on Security

1&1~=Umm • March 30, 2019 6:45 AM

On the Squid article,

“‘Stuffing an animal skin with a foam rubber replica—the classic method—doesn’t work for squids,'”

Did make me smile.

Use of ‘foam rubber’ is actually ‘the modern way’ of the ‘traditional process’. In times past, Plaster of Paris has been used as well as those good old furniture stuffing materials straw, horse hair, and cotton waste (all of which attract moths and other ‘wee beasties’ as kindergartens).

Plastification is very much not the traditional method and in taxidermy is concidered an entirely new ultra modern uncertain method that is not yet trusted. Because it derives from an earlier process where blood vessels were filled with a fluid that became plastic then the flesh was disolved away. There is thus the question of the ‘preserving factor’ of the original organic material and will it last three or four centuries or more that the ‘traditional methods’ can do.

I guess another example of ‘journalistic licence’.

1&1~=Umm • March 31, 2019 9:40 PM

@Tatütata:

“I forgot to mention that -0.8 is quite close to the reciprocal (-1) relationship of Zipf’s law.”

Which gives straight lines on log/log paper when plotted (as do a few other ‘laws’). It’s almost as though those straight lines are ‘natural’. Like the odd coincidence of the golden number etc, or fractal patterns.

However a couple of things that always give a crooked line no matter how you plot them is ‘military spending’ and ‘CCTV numbers’ ranked against either country or population… Which suggests in turn they are very unnatural 🙁

1&1~=Umm • April 2, 2019 7:11 AM

@Alyer Babtu:

“Suggesting once again that the only I in AI is the flat shadow of the intelligence of the human algorithm deviser”

I guess we should not realy be surprised, but then as a character in a black and white movie once observed ‘We expect such things from Frankenstein’s monsters godless soul’.

Deities aside, mankind appears to have changed little in 10,000 or more years physically, but what he can and has done is learn to not just live with the limitations, but learn how to use them to advance beyond the point of magic or whatever they could imagine even of those deities they could imagine.

The important point to notice though is that humans can not only feel pain, they can also know fear and death as an intrinsic part of their being. However as of yet machines are just not complex enough to get even close, so can not.

It is that intrinsic set of pain, fear, and death that is the foundation of not just our morals but our development beyond what we can do unassisted by our technology.

If you think about it an AI with physical agencie feels no pain on hard contact, feals no fear of future pain or death and has no understanding of what death might be. So what intrinsically within it is going to stop it hitting things with sufficient force to cause either repairable or even non repairable thus scrappage damage?

So we give it ‘rules’ but there is an old joke about those ‘All rules have exceptions’, they are by the logic they work on going to fail in some manner (see works of Kurt Gödel, Church, Turing and others as to why).

But rules are like laws an imposition of morals or simple dictatorship. They teach us to obay not to understand why they are needed not just for us but for others.

We have little knowledge of how humans learn, observations at best because our inate sense of pain gear and death tells us that actually opening people up and inserting probes etc is not what we would want done to us, so we should not do it to others.

There are two old sayings that pop up on most societies irrespective of contact,

1, An eye for an eye.
2, Do unto others as you would have others do unto you.

These only have validity when pain, fear and death have intrinsic meaning. As of yet we have not yet real world fielded A.I. devices with such intrinsic basis for learning.

It is not untill a child gets to around six years old does it start to realise that what it does has consequences to others, and about another ten years befor it’s thought they have sufficient understanding not to behave in risky ways behind the controls of a small vehicle. In the mean time we hope that they having riden scooters, skateboards and push bikes have learnt enough via pain and fear, to ‘not be stupid or atleast reckless’.

How do we get a machine to learn that way? The fact we can make a machine that can study to play chess in a day such that it can beat the best of humans, says we can get fixed rule discovery mastered but is that ‘learnt’ or not?

Humans supposadly have ‘free will’ that is on mass they appear to work by aproximate rules, but individually, no they can chose to follow rules or some interpretation of rules to a lessor greater extent. After all we still kill each other despite the rule ‘Though shall not kill’.

We think that ‘experience’ is what makes us ‘interpret rules’ and more importantly spot how others are likely to ‘interpret rules’ before they do, it is after all a primary survival skill.

What the experiment has show is two things,

1, Rules do not cover all eventualities.
2, Humans will exploit others weaknesses for their own advantage.

So how do we make rules to cover for things that we have not seen, our ‘black swans’ or ‘unknown instances of known behaviours’ or our more importantly ‘unknown instances of unknow behaviours’?

There are things that humans know, things that they can predict, but importantly things that are compleatly unknow to them. Rules can only sort of address part of the first of those and occasionaly some of the second. But what of the third, how do we deal with that? Or more importantly can we make a machine do that beter than we can?

I suspect not untill machines have learnt and understood pain, fear and death, to guide them.

Any way it’s getting to that time of day when I should be doing things other than typing round a cup of coffee etc trying to ‘saddle up’ for thr day ahead.

1&1~=Umm • April 3, 2019 2:05 AM

@Tat:

“But would you recognize “etaoin shdrlu if you saw it?”

It’s a ‘sin to er’, but yes I did 😉

But then I would wouldn’t I, I’m that kind of geek.

There is an argument about what order the letters actually are in etc, and it depends on your corpus and it’s place in time…

I tend to remember it via ‘Eat on Irish Lid” and strike duplicates and make two or three swaps (at,ni,ld) depending on the corpus nationality. I was puzzled by the,’r position’ in your list though. So followed the link

Yes it’s moved in around a century and a half… The sad thing is that I should have remembered Linotype… Their font rivals Monotype* had a factory in the UK built some time before WWII, in the 1980’s they built a new factory next door because of asbestos risk in the old building, then they effectively went bankrupt as their market colapsed due to DTP and Adobe products. I know this because for a brief 1 month period in the 1990s I worked in their new building. Why only a month, well I got bad vibes and left, in part because the day we moved in was April the first, and London was warm and sunny ‘short shirt sleeve weather’ when I left home that morning, at lunchtime at the new factory we had an inch of snow… The company that I had worked for that had moved in a little while after I left also more or less went the way of the dodo… Strangely though stuff I designed and did the software for back then is still being made and sold new, even though it’s a tenth the price…

One trivial thing to remember if you are looking at letter frequency in Ceaser ciphers it the ‘RST hump’ and the ‘AE gap’. It’s just one of those ‘fast tells’ when you are ‘working it by eye’, unless of course it’s Rot26, that just jumps straight out of the screen at you 😉

• The factory was in the delightfully named ‘Honeycrock lane, Salfords’ (not Salford which is ‘Hupt Narf’). There’s an aerial photo and a time line at,

http://www.eyemagazine.com/feature/article/a-monotype-timeline

1&1~=Umm • April 4, 2019 4:17 AM

India repeates others idiocy.

As has been mentioned on this blog before blowing things up in space is a realy bad idea due to Kessler syndrome*. Which makes it a real threat to communications security in the general sense and also mankinds ability to advance.

But for politicians and war hawks such considerations are a long way distant from their need to waste money and resources just to polish their egos, rattle their sabers, bang the drum, wave their flags and generaly behave like petulant six year olds.

https://arstechnica.com/science/2019/03/india-shoots-down-a-weather-satellite-declares-itself-a-space-power/

Mankind needs space to be usable for nearly all our modern information needs, and that becomes problematic with near earth space polluted by man made debris.

India like the Chinese back in 2007 could have shown more skill and thus ‘projected more power’ if rather than blow old things up they could have demonstrated a working ‘early deorbiting technology’. Thus what India has actually demonstrated is that they as a nation state are actually inferior to some commercial companies. Space X for instance has demonstrated it can not only hit a target in space automatically, but they can do it very precisely and safely certainly well enough to ‘lasso and deorbit’ another large object such as a decommissioned or failed satellite early such that it does not become an even bigger problem with time.

• Kessler syndrome / cascade is named after Donald J. Kessler a NASA scientist who in 1978, presented a scenario which he had realised and then did the math for to confirm it. He realised that the density of objects in low Earth orbit (LEO) was increasing and was then about high enough that collisions could and would take place between orbiting objects. Importantly that any such colision would result in rather more than the original two objects each of which had a higher probability of causing other collisions thus have a cascading effect. In a effect a run away or chain reaction could result where each collision generates nolonger functioning satellites as space debris that increases the likelihood of further satellites becoming non functioning. But worse it is effectively making the safe launch and deployment of replacment satellites increasingly less likely due to the very slow orbital decay rates –about 1000 years at 900Km up– of the accumulating debris. Why this happens is debris spreading, after a colision most of the debris is in a very limited volume in space, but due to slightly different orbital hights and velocities they will spread out over much greater volumes with time. Making this worse not all parts of a satellite can be easily found with radar / lidar and similar longer range tracking technologies. Currently the collision rate is about one every 11months but that rate is increasing, will this end up closing space to mankind? Well it’s upto us, firstly we can stop creating more debris, secondly we can clean up what is already up there whilst we can still safely launch systems to do so.

1&1~=Umm • April 5, 2019 8:36 AM

@Taz:

“Real or snake oil?”

Journalistic licence at best, more ‘click bait for the unknowing’.

The article starts with,

“‘Researchers have just released hacker-proof cryptographic code — programs with the same level of invincibility as a mathematical proof.'”

Which sounds grandiose, but actually does not hold water as an argument.

I’ll just outline some of the issues, others can fill in more 😉

There are numerous supposed ‘mathmatical proofs’ around and they usually mean squat diddly in the real world as you will find they are based on two assumptions,

1) It’s like another proof.
2) The assumption the other proof is valid.

Oh and the clasic ‘ignore reality’,

3) Scope of argument thus proof is constrained.

In effect ‘house of cards thinking’ where it’s assumed no breeze, table bumps, cats jumping up or other things not covered by the very limited scope will not happen and knock the house down…

If you look further down the article you will see the library is actualy based on ‘formal methods’. Which unfortunatly have more than a few assumptions behind them.

The first and most important is formal verification only veifies it’s input, not it’s output or what happens to that output, that is just ‘assumed’ to be secure as a given…

If people think about that a little bit, what formal verification is, is ‘A top down process that stops only just a short way down the computing stack’.

In many cases the output is effectively just more ‘source code’ thus it stops above the compiler pre-processor and all it’s subsequent parts to becoming running code.

As was pointed out long ago by one of the creaters of Unix, the compiler can be subverted in ways most, or even expert coders can not find.

But lets assume you are a real experts expert and can walk the assembler output –which is still just source code– and verify that?

What happens if I put my hack in the actuall assembler or linker or other object code libraries. Are you enough of an experts expert to walk through the actuall object code for each and every one?

But lets assume you are some kind of long snowy bearded guru steeped in loading your code with switches on the front panel, can you check the microcode inside the CPU?

Very unlikely unless you work for Intel, and since the Pentium bug, they patch the microcode everytime the CPU gets dragged into life, so ‘that’s a movable feast’ at the best of times…

But lets further assum that you are actually a hardware Guru with not just the beard, but a taste for real ale and a flat cap for the spreading bald patch the pony tail does not cover to stop the risk of sunburn if you ever go out in daylight. Can you check the translation to RTL and the underlying hardware macros used to build the chip?

Lets assume you can down to logic gate level, as you are some kind of computing diety with the abilities of thousands of others…

How about the device physics in every chip…

Security is as Einstien observed about the universe “reletive to the observer” and also bound by not just the observers scope but by the universes rules as well.

But those last bits about ‘below the CPU ISA’ are in no way theoretical there are practical attacks such as Rowhammer and Meltdown plus there was also all those I/O attacks due to DMA and similar on MMUs that popped up from time to time since atleast the 1980’s to my knowledge.

But there is another problem worse than ‘Black Swans’ which atleast are attacks possible to envisage, there are without doubt attacks to be found that have no commonality with other known attack classes, thus can not be envisaged based on existing knowledge…

So is this software library secure?

Not untill all the above issues have been fixed… That’s not to say the library is insecure but you can not prove it is secure in practical implementations, and that at the end of the day is why a lot of money is spent on EmSec (TEMPEST on steroids) and ‘segregation’ via SCIF’s etc. Or as christened on this blog ‘Energy Gapping’.

I hope that answers your question to your satisfaction (and others reading along).