Peter Bärnreuther - Taking on Slashing Risk: Slashing Risk from an Insurer point of view

Transcript

yeah last speaker of the day Peter the floor is yours thank you I think I have to applaud you for still being here meaning that your staff are really interested into Insurance yeah one of one of it is it so welcome to the talk taking on slashing risk from insurance point of view um my name is p bandr I'm the chief underwriting officer of chamb brof um um let me shill Cham in one slide we are fully regulated insurance company so picture S as a traditional insurance company taking on web free risk we licensing Bermuda we have wonderful

Partners like Soo which is um japanes second largest insurance company Munich re largest reinsurer in the world and Guan stem leader one of the leader in smart contract or general web free security and but enough about that that we are talking VI are ensuring against smart contract and slashing risks so we are often talking to node operators about slashing risk and we saw that we have very different perceptions um what node operator how node operator see the risk and how insurance companies um see the risk and I'm not talking here about um about uh chain proof but I'm talking about the whole insurance industry so do

did slashing event happen yeah 300 slashing event happened on the main l in the last three years that's not too many if we say we have like 800 active validator let's be fair perhaps in the last three years on average it's more like 200 validators um the right um annual insurance rate should be more like should be you can argue like 330 e which got slashed divided by three because three years and then divided through this St L East which is 32 East times 200 validators and you come up with a very very low rate so even if you're then a midsized uh validator you shouldn't pay more than one or 2,000 um

2,000 um US dollar per year now you can imagine that if you call a insurance company I want to insure against a slashing loss of 100 million but I'm paying perhaps or I think the fair price is 200 uh is 2,00 the contract probably will not happen so why is this or how do insurance companies think about this risk so basically how to price a risk there are two ways first way you have a lot of data then you just use um then you just use the data of the last perh let's say 50 years and adjusted a little

bit like with car datas uh um traffic data you know the value of the cars going up the driver Behavior goes a little bit up but basically you have data of 50 years so you can really really do your pricing but what is about new risk yeah three years is not a long time so how do we price three risk and one common method you take you're using is basically you split it up in three different parts in basis losses these are the very small losses like one one node gets slashed might it be from a large node operator or solo Staker that's to but that's totally irrelevant from a cost P perspective perhaps not

from a reputation perspective both for the insurance industry and for the node operator then we have large losses this in an insurance industry would be the mean that perhap the whole node operator gets slashed so we are talking here about six seven uh figures so it's really relevant for the node operator this might be a huge operational failure or and as well uh relevant for the for the insurance company so for both and then we have this catastrophic losses where basically the whole isum network goes down or is threaded to go down but where from insurance perspective not only one node operator um has losses but

several of your policies at the same time default and so you your losses adds up these is basically the the mental model which which we are using so basis losses we have seen quite a few but large losses and catastrophic losses never happened or did they I think they did so let's let's put in let's go in there I brought eight cases with me um where I think large losses or catastrophic losses almost happened um and and let's go through them so the first one is is a a software buck in the prism client so all perhaps all of them

um you can find on chain proof Point Co it's co.com at our blog um the lest blog post I describe exactly these eight um accident and as well with um you find the it's a the original postmortem so there was a wrong response of one of six different time servers so this led um to a event where the prism blind um was not uh thought it was a different uh time than he he saw the stems on the beacon chain so basically no attestation was more going on the prism team rushed them to push a fix to the problem unfortunately this fix U contained then a critical buck and 3,000

validator has slashed luckily on the main net H sorry on the test net so nothing really happened but you know something like this totally unexpected might come up one time in the main it as well second example I think is a classic node operator node operator drives to to um to get a better performance he's testing a test net main depends uh behaves differently so the no start some validator nodes start to restart the um slashing prevention database is disabled never a good idea and so there were a lot of slashing

events around 75 or or 95 there are always different number around um of these notes for slashed so one 100 e for lost would be now about 200,000k and at this time uh the validator count was quite low so um I would argue this is already a u yeah let's say me medium event then example three again a soft back in our valid dat client um they were un unable to produce blocks in a period of 18 epos and again 18 epos this led to a decrease in validator participation um a league mode was not

triggered but it was very close at this at at this event example four um a note operator had a lot of valid dator clients several beakon nodes but all valid dator clients were um basically connected to one beak noes um due to an operational failure and uh because all of these were hitting a single beak node it become much mess less responsive and some blocks were becoming oround so this is perhaps not a slashing event but um where you can see okay an operational failure is ongoing um example five it was on a test net

here A F created invalid block and this was deemed valid by two of the execution client and so the the network split up yeah if you would have this on Main net three the the main chain splitting into three different one this would be catastrophic um example six um This was um basically uh a a the prim and primary um net was on and uh and um the node operator deleted some of the validata clients but they it wasn't

really deleted the keys were reimported so suddenly you had uh backup validators going online and this cost a slashing you one from 11 East then example seven um some consensus client could not proc it valid attestation with an old Target checkpoint in an optimal way so the the network was flooded with all attestation and I think three out of the four main clients H recomputed prior Beacon states which was a lot of computational power and the network went down so we were in league mode because only and basically the network was saved by one validator

client but the network was in league mode um since we have the quadratic formula the impact was very small but nevertheless and then another event only like um one month ago 20 note were slashed because the fallback data center was brought On and Here the web resigner didn't had a slashing protection a so these were um eight cases um so let's review our our basic assumption again we have the basis losses here I would say we have already like 150 events have happened we would have already some data

but unfortunately nobody wants to ensure against this because it's just irrelevant from a monetary perspective so let's delete these and the big question is now about large losses and catastrophic losses yeah large losses we have seen can happen this would be an operational failure like the backup system goes on and again we haven't seen it yet but in principle an external hack can can bring down the infrastructure or steals the validating key and then enforcers a slashing same could happen and crumpy ex internal employee who still has access or some software bucks so

four of them happened in the past the severity was low luckily but this might be might not be might be that it's not this case in the next 100 years um and and similar and I would argue we had some more I haven't seen for everybody for every one a postmortem and what is with this catastrophic losses if you're making a little brainstorming there might be a valid dator buck ex execution blim Buck we have seen both of is okay a failure of the underlying ethereum Network after an upgrade might come DVT and overloading consensus I still have to make my mind

up so not sure about it and um leag mode I think is is very realistic scenario as well so the question really boils down and this is where the different perception is coming in insurance is not against these basis losses yeah we have seen these on the network but nobody buys insurance for this one this would really correspond to this low rates from from 0.002% the real question is what do we give the probability for large loss or for a catastrophic loss and the answer is we don't know yeah we have to make some assumptions but basically here here it's important you think it's one time

every 100 year two times every not year and then you can come up with a um basically Fair rate so in the in the end what can we do from the insurance industry um well um or what is our challenges well it's very technical it has quite a technical complexity even for a cyber underwriter and the data quality is quite okay but the quantity is not there yet and a challenge for us is as well the um accumulation risk so um leak mode downtime software box failure of the underlighting Ling eum Network this

would be risk where perhaps many policies separate are affected and this would perhaps lead to a large loss which we as insurance company have to have to avoid and um how to overcome these challenges um mitigate risk and get insurance cost down what can we do um well the data will improve over time so we get this in our favor we can do technical improvements DVT might be a solution I think we still have to to wait for the results from the test Nets doing a little bit more research here what are Insurance specifics we can try to exclude these cat risks to make it

insurable for us and uh what we are always using in the um is these align of Interest through safe insured retention and loss participating so that we both have an interest that not slashing occurrence then last Point what can node operators do way doing an awesome job at the job at the infrastructure setup of course education of the insurance Market is always a topic as well because it's quite difficult what what you guys are doing um following best practices uh um like um we which we be doing currently together with Lio and the duck

initiative and then ISO 207 27k and sock 2 certifications um are also I think something what a lot of node operator is doing currently just to give these positive signaling effects and last not least qu them is doing specialized slashing specific infrastructure audits which are basic basically complimentary to the iso and sock 2 um certifications that's from my side thank you for the that you have been here um if you have any question you will find me outside or feel free to shoot me an email if there any thank you and if there are any

[Applause] questions questions maybe I have one um can you Ure against lost keys this would be custard risk um there are many we and chain proof not doing it but there are many companies who are doing it this is more a TR a risk which traditional insurer feel more comfortable so like coinbase also they they buy a lot of of these risks um hot wallet risk or cold wallet risk both of them yeah hi thanks for your talk uh I'm just

wondering is this a market for for institutions only in terms of risk mitigations and risk policies or is it also like is it for uh solo policies let's say for solo staking is there is there demand or do you offer product products for this I I don't know if there's demand but it's difficult because the insurance companies always do some risk assessment and with solo stakers is just tough yeah so usually it always depends what is the technical knowledge of the insurance company if it's low then they rely on certifications like 70,000 K audits ISO 70,000 k um certifications and then basically it's

getting tough for solo stakers um yeah so solo stakers with few nodes then the premium would be as well too low for the insurance industry or to the insurance company to start the process so thank you great talk um my question is what drives the demand of the insurance like what kind of people are coming and do you see during the last years that the demand of slashing insur insurances are increasing wec basically node operators who talk to institutions um they are asked by the institutions by the compliance team of

the institution do you have insurance and that's a checkpoint and if you say no and three other SN operators say yes you're out of the game so basically it's um yeah it's an sign it's a marketing tool a signaling tool and a risk transfer tool and um what we as well see node operators are buying it when they think they get more business out of it so that it basically pays for itself okay thank you [Applause] again