Things I wish I could do with AI (that is still surprisingly hard) 🧵

I looked into *why* the supply of Covid-19 rapid tests has been so badly disrupted in the UK. TL;DR: All of the distribution is being handled by a single company which will be closed for 7-8+ days around Christmas and New Years' Eve and therefore shortages will persist. 🧵👇🏾

"Nobody could have predicted the enormous scale of the surge in demand for COVID-19 test kits that we saw yesterday, following the Prime Minister’s message to the nation on Sunday night." NOBODY could have predicted?! 🤔 psnc.org.uk/our-news/lateral…

Given that the govt/NHS should have known back in the beginning of December that their distributor was planning to close down for a significant portion of the holidays when there would be high demand, it's pretty shambolic that they didn't have backup distribution plans in place.

The fact that the award is a single company leaves zero resiliency in situations like this, when that company (for legitimate business reasons) needs to close due to staff availability. Everyone deserves a holiday. But this meltdown in test distribution could have been foreseen.

All of the distribution for Covid lateral flow test kits is being handled by a company called Alliance Healthcare along with its logistics arm Alloga UK. Here are the list of holidays/closures it announced on 1st December across England/Wales, Scotland, NI.

54-59 packs per day, per pharmacy is an insanely low figure since this is the default/recommend method for collecting LFD kits. Statements from the govt that focus on overall numbers of tests made available are a red herring, since people collect these as packs of 7, not loose.

Distributions only resumed on 29th Dec and will be closed again 1-3 Jan (1-4 Jan in some UK countries). Overall, there's only approx. 1.25mn Covid LFD test packs that will even be processed at the distributor. (There are ~27.8mn households in the UK.) psnc.org.uk/our-news/pharmac…

A large pillar of distributions for rapid tests is the Pharmacy Collect service, rather than home deliveries. Pharmacies can order a max. of 1 carton (with 54-59 packs of tests) with a max. of 6 cartons per week. Each pack contains 7 tests.

UK Health Security Agency (UKHSA), the health ministers etc statements on this situation (not a direct quote) reminds of this evergreen article from The Onion, which they post after every gun violence incident in the US. theonion.com/no-way-to-preve…

@zsk @BBCHughPym @KevinJRawlinson @BBCRealityCheck @GdnHealthcare you might find this thread interesting on Covid test availability 🤓

Other companies involved: 1. £550mn to Royal Mail for home test kit distribution (up from original £270mn): find-tender.service.gov.uk/N… and find-tender.service.gov.uk/N… 2. £200k in a previous LFD distribution contract to Pheonix Healthcare: find-tender.service.gov.uk/N…

Charitable explanation: police are used to / taught how to look through footage for human suspects, where a binary search type strategy won’t work (suspect might only be on 1-few frames). So they might be habituated to always thinking “need to watch whole tape”. Depending on what the IT setup for CCTV footage is, if it’s designed always to play/fast forward/rewind, it might be very cumbersome to seek to halfway mark. I’m thinking of airline in-flight entertainment where there’s no simple seek to time or drag to time option, only a very cumbersome and laggy fast forward which is fiddly and annoying to use. The actual issue of course is even if those barriers didn’t exist, maybe they aren’t incentivised or promoted for recovering lost property.

There used to be a difference between the two, but since it caused so much confusion there’s no official difference any more in the W3C specifications. url.spec.whatwg.org

Top-loading washing machines consume far more energy, as all the motion/agitation has to come from motors. Side/front-loading washing machines only need to use a little bit of energy, and then gravity does the rest.

Sure, but the other distributions to home are done by Royal Mail which is also closed during the holidays, and to community centres is also handled by Alliance Healthcare as they handle warehousing and storage.

I love the moment of delight when apps email you back about a bug that you reported getting fixed (even if it takes 2 years to fix the bug). This specific one is from @Flighty, one of my favourite apps. ❤️ Actually going and tagging user reports to a bug report and tracking it through to completion is something not a lot of apps/services do, leaving the user guessing on whether it got fixed or not. And did it get fixed in exactly the way they wanted, or some combination of feedback from different users?

This partnership has been a long time in the making and I'm so excited that we finally announced this. Being able to partner with Timpi is huge for @cheqd_io's #VerifiableAI roadmap. For some context... Training any AI model requires mountains of data. In practice, this runs into 100s of TBs...which means the only companies that are realistically able to do it are often the Big Tech companies that have billions of dollars to throw at this problem. Which is why you'll see all frontier models come from players with extremely deep pockets, such as Google (which already indexes the web), OpenAI with Microsoft Bing, Meta/Facebook etc. (Google is one of the sneakiest among this, btw. You can't opt out of having your site indexed for training their LLMs, without also opting out of being included in their search index. Not being included in Google search results is a kiss of death for any traffic, and so people begrudgingly stay opted in. This is a kind of abuse of market power that I suspect will attract the attention of competition regulators soon - or maybe Google's strategy here is to do a landgrab on data as fast and quickly as possible before the two antitrust cases brought against them breaks the company up.) Once you train a model, for "fresh" results beyond the training dataset date, once again you need a search engine feed. This is why ChatGPT used to say "according to my training data till Oct 2022..." and only recently has had capabilities at the free tier where it can plug that gap with Bing search results. Why does this matter? If we truly want decentralised AI, we can't just be running the scraps of open source models released by Big Tech (and quite prominently, neither Google or OpenAI open source theirs anyway). Repeat after me: decentralised AI needs decentralised training data. And as it happens, getting this is incredibly hard...and why we're so pumped to be working with @Timpi_TheNewWay as they have solved exactly this problem, at scale, with one of the largest search engine indexes after Google/Bing. So where do digital credentials come in? I can foresee at least two paths where @cheqd_io's identity stack comes into the picture: 1. Training data sets for AI (narrow OR broad) will shift from being distributed via centralised servers/channels (like they are now) to decentralised distribution. AI developers will need to validate that these datasets are untampered and understand the reputation of sources in the dataset at a granular level. Verifiable credentials - with privacy-preserving payments to the sources checking the quality of the data (in this case, Timpi/cheqd) - are a solution to solving this. 2. As a greater proportion of the content on the Web is AI-generated, model training starts getting worse and at some point entirely collapses (at levels as low as 5-7% of synthetic data). There are two ways Content Credentials come into this: - If/when Content Credentials become commonly-attached to content (say 2025+), search engines will need to parse these Verifiable Credentials as a signal to understand how much of a web page is AI vs human generated. - Even when Content Credentials become popular, it's not guaranteed that every piece of content will be accompanied by one. So similar to general quality of web data, one of the aspects that could be encoded into are Content Credentials (or Verifiable Credentials) are heuristic assessments of whether the site/content is AI vs human generated. In both scenarios, we at @cheqd_io believe payments for credential exchange can be incredibly useful for implementing payments for royalty/licensing checks when training AI models. Extending these ideas further, as agentic AI becomes viable, AI agents themselves would need to prove what sort of quality of data they have been trained on, compliance with various global regulations, and using Verifiable Credentials as a mechanism of show what permissions they'd been given agent-to-agent or human-to-agent. As we said in our original hypothesis for $CHEQ, MANY of the credential exchanges/interactions would require a payment mechanism to accomplish value exchange. We're starting to see some early implementations of this, i.e., AI agents paying each other using traditional fiat payment rails like Stripe...but I fully believe that for this technology to scale, in practice they will require decentralised, privacy-preserving payments for credential exchange for #vAI. Excited to build together with the @Timpi_TheNewWay team towards this vision!

Privacy is a bit like exercise. Everyone wants more of it, but actually getting it requires work that feels like homework - and nobody likes homework. 📚 This absurdity becomes clear when you look at examples like a cookie consent screen with over 1600 choices. Even if you took ONE SECOND per choice to think about whether you wanted to approve or deny, that's nearly 30 minutes of work. The model of "privacy-through-endless-choices" is fundamentally broken. It puts the burden on the user, and it isn't working. It reminds me of 2000s/2010s internet: we used to tell people, "Look for the padlock icon 🔐 to be safe." The responsibility was on YOU to check. That advice has now disappeared, NOT because security is less important, but because GOOD security (HTTPS by default) became so ubiquitous that the infrastructure made itself invisible. We've seen a clear technological evolution: First, we encrypted data-at-rest. Then, with TLS/HTTPS, we secured data-in-transit. The final frontier has always been protecting data WHILE it's being used. Privacy technology should protect users, EVEN IF they do risky things with their data. This is where the paradigm shifts from privacy-through-choices to privacy-through-architecture. And this is why I'm thrilled to announce that I've recently joined @zama_fhe. 🎉 Zama is at the forefront of turning the most advanced cryptographic research - Fully Homomorphic Encryption (FHE) - into practical, usable technology. FHE is the key to that final frontier, allowing computation on data while it remains *fully-encrypted*. 🤯 This could translate to a vision for apps where: ❌ No more consent screens ❌ No more privacy homework ❌ No more asking users to become experts at reading Terms of Service ✅ Privacy protection happens automatically in the infrastructure layer - users get the benefits without any cognitive overhead My mission at Zama is to help productise this vision to take this revolutionary technology and shape it into tools that make privacy an invisible and effortless foundation for the applications we use every day. We're still SO early in the journey for a more private-by-design Internet. If you're passionate about solving fundamental problems in privacy and baking it into the future of the internet, I'd love to connect. 💪🏽🚀

The irony is that they work with 3 major CDNs: Cloudflare, Akamai, and Fastly for iCloud Private Relay and all 3 of them could perhaps provide much faster download speeds than this.

Super proud of the team at @cheqd_io for being considered in the top 100 🔥 Next year, we're gunning for the top 10💪🏽

Exciting to see @British_Airways trialling digital @DecentralizedID credentials for faster, simpler international travel clearances, using private and secure W3C #VerifiableCredentials. (This is the same use case that @fraser_again worked on at Known Traveler Digital Identity.) More and more examples of digital credentials appearing in the wild 🙌🏽 forbes.com/sites/davidbirch/…

Something veeeeeery interesting happened last month on @cheqd_io's testnet: we suddenly shot up from approx. 18k DIDs to 70k+ DIDs, and 10k+ DID Linked Resources. We're not quite sure who did, but this to me is a tell-tale sign of someone running a performance test, where you fire off requests at a rapid pace to simulate whether a software can handle the volumes necessary for large-scale production use cases. The absolute numbers themselves aren't very relevant, but the rate of change. And I'm happy to say that as far as the network was concerned, we kept chugging along without any issues - didn't even realise this took place until a few weeks later when looking into product stats! And this is likely not even pushing the system to its real limits. Performance tests come in many shapes and forms: 1. Ramp and hold: Slowly ramp up the rate of requests, e.g., from 1 per second to 10s per second to 100s per second and so on. At each "step" you hold at that rate for a defined time, say 1 hour or 2 hours. 2. Soak tests: Fire off a high rate of requests, say, 10-100s per second and sustain it for a long period, e.g., 24 hours, 72 hours etc. From the data analysis, we think this was a soak test running over multiple hours. If you're the developer(s) who were working on this, do get in touch with us! We'd love to chat to figure out what use cases you're working on and what sort of performance you require.

A cautionary tale about why we need ✨ Verifiable Reputation ✨ when hiring remote developers (because they could VERY LIKELY be North Korean state actors)... You might think I'm being overly dramatic, but once you see the evidence laid out in this thread, it's not THAT big a leap to believing these were North Korean state-sponsored hackers, given this is a known modus operandi for the Lazarus Group. More in the thread below 🧵

🚨 We built a network-wide validator status monitoring tool for @Cosmos SDK chains that can monitor for validators that have missed too many blocks, and are likely to get jailed. Check out the code on @cheqd_io's GitHub 👉 github.com/cheqd/validator-s… Here's how we went about it...

After an incredible 4+ year journey co-founding and serving as CTO of @cheqd_io, the time has come for my next chapter. Earlier this week, I started transitioning from day-to-day responsibilities to a role as Board Member and Strategic Advisor, and I couldn't be more excited for what's next. This isn’t goodbye. It’s a natural evolution reflecting our company's maturity and the strength of our team. Building cheqd with this team has been one of the most rewarding experiences of my life. As a remote-first company, we built a vision and a team entirely through screens. That’s why our first all-hands trip to Istanbul was so special. Finally meeting in person was electric. My favourite memory isn’t a late-night brainstorming session, but the chaotic spicy noodle-eating competition we organised in a hotel lobby. 🍜 🤯 🥲 Those moments of shared laughter are what truly build a team. That sense of fun was core to our culture, from virtual escape rooms and Among Us socials and go-karting and VR zombie hordes, to commemorating our first mainnet DID with a Star Wars-style movie poster. We always found ways to connect and celebrate the wins, big and small. I’m so proud of what that team achieved: launching a network, pioneering DID payments, and driving real-world adoption. I know our team is ready for what's next. As we enter the Era of Hyperpersonalisation, where nearly every product and service becomes AI-powered, the urgency of our mission to protect data, give people agency over their identity, and build trust only grows. I leave the day-to-day confident that our team will champion this mission with passion. 💪🏽 🚀 In that spirit, I'm sharing a side-by-side: a photo of Fraser and me from when this all began, and an AI-generated, Studio Ghibli-style portrait. The contrast shows how fast the world has evolved, but the core of this journey remains the same: starting something real with people you trust is a legacy I'm proud of. In my new role, I look forward to supporting cheqd’s continued growth and success. To my friends and colleagues in the decentralised identity community: thank you. I've cherished building alongside you. My day-to-day focus is evolving, but my passion for our shared mission is unwavering. I’m not going far and hope our paths cross as we continue building this ecosystem. Deepest thanks to the cheqd team, our community, and especially to @fraser_again, @javkhattak, and @EHotta. Your partnership and friendship have meant everything. ❤️ Proud of what we’ve built, energised by what comes next for cheqd, and thrilled for my own next adventure: immersing myself in the vibrant deep tech scene here in Paris and focusing on advanced privacy technologies. More to come!

I've been a judge at many hackathons that have 100s of participants, and many of them use a pairwise-voting system called Gavel or Jury (Gavel was created for HackMIT, which has 1000s of participants). Basically, the idea is that judges ONLY answer "is the current hack better than the previous one?", that's it. The software then sums up votes from judges and uses a statistical model to create a ranking/distribution of all entries. It's much faster and simpler than having complex spreadsheets and asking people to assign a score in different categories. Gavel repo: github.com/anishathalye/gave… Jury: github.com/hackutd/jury

Both @creds_xyz and @discoxyz getting a shoutout in this report on decentralised identity from #onyx by @jpmorgan 😎. A good, balanced read on the challenges with NFT-based POAPs 👌🏽jpmorgan.com/onyx/documents/…

More on this from Sajid Javid basically confirming along the same lines that there are tests on order, more in the way, but distribution is a challenge. inews.co.uk/news/why-cant-i-…

Earlier today, our X/Twitter account @cheqd_io was compromised to link to a phishing website. This was an *incredibly* sophisticated social engineering/phishing attempt, where many of the usual red flags were hidden expertly. Wanted to share this as a cautionary tale... 🧵

Excited to see this flag planted by Ethereum Foundation to double down on privacy. A huge cluster of different privacy technologies that will shift the paradigm on what good looks like. 💯

It’s hard to understate how tectonic this shift will be for how our personal data is handled online, if the DOJ indeed gets its way to force Google to divest Chrome into a separate company. Chrome is critical to Google’s dominance in the search engine and adtech business. As a separate company, Chrome would no longer have the unlimited wealth of Google’s billion dollar ad business to tap into. And even for browsers like Mozilla Firefox, a lot of their revenues come from being paid to have Google search as default - which the DOJ is also asking to restrict. In such a world, Chrome might finally be forced to innovate using ad models that rely on “zero-party” data - ie, preferences and data directly revealed by users - and have to compensate them using the kind of value exchange mechanisms we’ve built at @cheqd_io. theverge.com/2024/11/20/2430…

Home deliveries are indeed done by Royal Mail and apparently they’ve now roped in Amazon too. When the pandemic first started, there was an emergency direct award for PCR test kit distribution that was handled by Amazon circa May 2020 and branded/dispatched via Amazon Logistics.

This has been in the works for some time, and we really love the @docknetwork team and product. From a product perspective, it's one of the best and most mature technology products in the decentralised identity space. I often take Dock Certs as an example internally of what a good digital identity product looks like. I have a lot of genuine love and respect for the Dock team on that. Our vision has always been to run a big tent approach and try to bring in as many developers on our platform as possible. There will be work to do here, but I'm genuinely excited that Dock and its customers will be able to switch over to the cheqd network's identity functionality. At a technical level, this has been possible because of the early thought we put into concepts such as DID-Linked Resources, which gives cheqd the flexibility to support bringing across developers like Dock and all of the use cases they would want to build.

Excited to be representing @cheqd_io and joining the industry-leading folks on the current @DecentralizedID SteerCo to help shape the future of a more secure & private Web 💪🏽

It's kinda crazy how fast the MCP specification (for AI tooling) is evolving. We've been working on building out MCP server plugins that allow AI agents to read/write from @cheqd_io (close to being released soon!) - and within the last assessment, a week or so ago to today, some features that weren't possible (like remote MCP servers) have already changed! Right now, MCP servers are designed to be running locally, say, on Docker containers. This is fine for developers, but obviously limits how many people can use such a plugin - with remote servers, you can just consume it "as a service". Would be a real game changer to accelerating adoption. (The other bit I'm curious about is whether Google, OpenAI etc also embrace MCP.) Good overview of the MCP stack and value it brings here: a16z.com/a-deep-dive-into-mc…

Thanks to the all the partners and validators on our network, we had consensus restored in approx. 15 minutes on this upgrade to v3.x. We’ll continue working with the remaining validators to help them upgrade. There’s a lot of ideas we have cooking for the roadmap for 2025. This will be the year of decentralised identity x AI. 💪🏽

Building @creds_xyz over the past six months is our effort to make decentralised digital identity *real*. We’re doing this by trying to make social reputation portable across Web3 online communities. We want to show that SSI can demonstrate product market fit by showing off a north-star use case that is *fun* and *engaging* and can have 100k’s/1m’s Monthly Active Users. I’ll be running a deep-dive session today at #IIW explaining how and why we’ve constructed creds.xyz as an open reputation system. Come find me if you’re here! @rosspower has done a much better job explaining the deep product research our team has been doing across the Web3 ecosystem. (Registrations are currently on a waitlist as we slowly open up to beta testers over the next few weeks.)

If he was travelling on an ESTA, “working remotely” is explicitly not one of the things allowed. (Many people get away with it by not mentioning that’s their intention.)

For a deeper, technical dive into @Creds_xyz, come join the @cheqd_io team in the Panorama room on how to join our beta for issuing reputation credentials on @cosmos 🤓🧑‍💻

Hey all, you might have noticed that the @cheqd_io validator is down/not producing blocks. Please don’t be worried: this is a controlled shutdown we’ve done to slow the network down for the next 24 hours. Here’s some more context: - As you know, we have a software upgrade proposal active for our mainnet upgrade to v2.x bigdipper.live/cheqd/proposa… This proposal ends its voting period around 0940 UTC. - When you put in a software upgrade proposal, the “time” the upgrade is meant to be executed is set as a block height. Usually, we calculate and set this to the estimated block height at the proposal end period + 1-2 hours worth of additional block height to give some buffer, for network block time variability. - We did our block height calculations based on approx. block time of 6 seconds, which is what the block time was last week/leading up to last week. - Unfortunately, after the weekend, we found the actual block time on our network was averaging around 5.6-5.8 seconds, i.e., faster than we’d anticipated in our calculations. The upshot of this would’ve been the block height would be reached 10-15 minutes *before* the voting period for the proposal is meant to end. - If the upgrade height in the proposal is reached before the voting period for the proposal ends, in effect the vote is cancelled as a null vote. If the vote gets nullified, we’d have to put in another proposal and vote on it + wait 5 days before the upgrade could take place. - So, to slow the network down, we’ve taken down the cheqd validator (since it’s one of the large validators). This slows down the block formation process as it takes the remaining validators a bit longer to reach consensus. Indeed, this has increased avg block time to about 6-6.4 seconds, which will ensure the upgrade height *is* reached after the voting period ends. (Slightly slower than 6 seconds is fine, as it negates some of the period during which the network was running faster.) - We’re keeping an eye on the validator to make sure it doesn’t get jailed. Jailing due to downtime only occurs if the validator is down for 20+ hours, so this criteria won’t be triggered as this is controlled downtime. TL;DR: You don’t need to take any action, such as undelegating or redelegating from the cheqd validator. We’ve done a controlled shutdown on it temporarily to slow the network down a tiny bit, so that the software upgrade vote/proposal doesn’t get nullified.

Fun to meet our friends @AnimoSolutions (@TimoGlastra , @ssi_karim) and @veramolabs, and many others not in the pictures at #IIW. So much positive feedback for @creds_xyz and everything else we’re doing @cheqd_io wouldn’t have been possible without the support of partners like them and @walt_id ❤️

(I had screenshots and thread on this in 2020, this for example is the screenshot of a dispatch notification via Swiship.co.uk - an Amazon Logistics website - for PCR tests. Will try to dig that thread out.)

Nice to see this acknowledged at EIC by EBSI! Making payments associated with credential exchange privacy-preserving is a challenge we've spent a LOT of time getting right @cheqd_io, and we have the toolset to go do this in production, right *now*.

There’s obvious things like name and address that might change (albeit very infrequently). The most relevant reason why it’s refreshed is because it’s possible to report a driving license as lost or stolen, in which case the physical document still exists but is no longer “valid”. KYC systems (at least the good ones) will do a live check against passport / DMV APIs to check whether the document being used is among those flagged by law enforcement, or reported as lost or stolen.

“People need self sovereign identity to truly have ownership in this new world we’re building” - @jessepollak Coinbase spent about a year looking into so many different aspects of identity. But…identity is a value prop is hard to sell within organisations, which is why it takes so much time to build things in this place.

Exhausted from a trip to/from Canada - my first international trip since my ski accident - right after the Olympics, but probably one of the most thought-provoking workshops on decentralised identity I’ve attended in a while. Can’t go into too much detail just yet, but I haven’t seen this level of cross-industry alignment in a while. 🤓

Got a quick picture of all the sessions happening on Day 1 of #IIW @idworkshop. (Twitter only allows posting 4 pictures in a tweet, so session 5 slots follow after£

This has been a long labour of love to put together. 2022 was about building foundational basics of decentralized identity: easy-to-use SDKs, innovations such as DID-Linked Resources (DLRs) and AnonCreds on cheqd 2023, we shipped one-of-a-kind payment rails for digital credential exchange; Credential-as-a-Service; and @Creds_xyz making your Web 3.0 reputation in your control and portable, secure, private 2024 will be about next-generation cheqd network with improved tokenomics and liquidity; better SaaS tools with support for emerging standards; a broader vision to take Creds.xyz beyond just reputation; and experiments in how this tech could be applied to emerging areas such as artificial intelligence, online personalisation, and more.

Proud to have @cheqd_io as one of the sponsors for the @DecentralizedID Hackathon running in October (and nice to see some familiar faces like @AnonyomeLabs and @vidos_id as sponsors as well!) We're excited to set our challenge as Harnessing Decentralized Identity for Verifiable AI, as we see this as an emerging area of risks and opportunities such as Content Credentials, credentials for dataset quality, proof-of-personhood, and proof-of-approved-AI-agent add more trust in an age where AI is ever more present. We have two prizes: $5,000 worth of $CHEQ tokens for the best entry $2,500 worth of $CHEQ tokens for the runners-up entry

"Sarawut Sanit" was one of the North Korean threat actors whom we rejected from the hiring process when we realised that he was likely a hacker trying to infiltrate our company. @skesslr Love the reporting! You might find it interesting to read some of the tell-tale signs that we discovered @cheqd_io that sets such North Korean hackers apart 👉🏾 nitter.app/ankurb/status/18038556…

Super proud of the team for this (last year, we ranked 60th!) 🥳

Here’s the announcement that was done in April 2020 about Amazon handling PCR kit distribution thesun.co.uk/news/11417342/a…

Congratulations to "vAI (Verifiable AI)" chatbot on winning this prize! In terms of sophistication and UX, it was one of the best entries we saw in the @DecentralizedID hackathon. The chatbot uses DID-Linked Resources (DLRs) using cheqd Studio - which I'm sure would make @Tweetddale happy! - to publish chat summaries, which would allow for checking whether a bot truly gave that particular response or not. It's obviously in very early stages, but some modified form of this could be important for proving human-to-agent or agent-to-agent permissions. I know @EEAnder1 was quite impressed by the technical quality of the submission too. Now...for the ugly part: many submissions (generally, not just our challenge) in the hackathon were quite clearly AI-generated slop. You could see that the codebase was doing something entirely different than whatever answers the submission gave for "How does this use DIDs and VCs?" The ease with which people can generate slop to submit to hackathons using generative AI is certainly a challenge hackathon organisers will need to start dealing with. You might even want to gate entries using some form of decentralised reputation, to capture previous wins OR mechanisms to say "this submitter was previously flagged as submitting poor quality entries" from hackathon organisers.

Super excited to finally be announcing this. If you haven't come across it, "MCP" is the emerging de facto standard (within the last 1-3 months!) for how AI apps/agents/tools talk to each other. In simple terms, say you wanted your AI agent to have access to Google Search, it would use the "Google Search MCP tool". Similarly, if an AI agent needed to understand and use decentralised identity, it would need an MCP tool - like the one we have just released - to understand and use it. One of my big goals for @idworkshop is to make more @DecentralizedID ecosystem to make decentralised identity a first-class citizen for AI. The big opportunity here is that I see every agent will need an identity wallet to store and share accreditations, and this applies for agent-to-agent, agent-to-tool, and agent-to-human relationships, which is far more in sheer numbers than just human-to-human interactions. This is not the end of the road. MCP as a specification is rapidly evolving, and doesn't quite handle anything identity yet. Honestly, what was super cool is to tell Claude Desktop to issue a zero-knowledge credential, and it automatically figured out that it needed to create a DID first, and then published additional files on our ledger to make the issuance possible. Without MCP, it would have not known how to do that, or how to read decentralized identifiers. Watch this space, we have more things cooking here on how agents can prove they are trusted. 🤖

Come find me to talk about: - Standards in @DecentralizedID and what the Decentralized Identity Foundation is doing in this space - “Verifiable AI”: how Content Credentials and decentralised ID apply to AI use cases - AnonCreds, selective disclosure credentials and privacy-preserving payments for identity exchange work we’ve been doing at @cheqd_io

Just verified my profile on @LinkedIn, which is one of the early adopters of decentralised, portable, and secure identity at scale. Unlike Twitter/X, where the verification is just a proxy for "do you have a credit/debit card?", LinkedIn gets this right by using trusted data sources (in their case, scanning the chip on your passport). Interesting design decision here to not make the verified tick too prominent (i.e., no prominent "blue tick", just a subtle status indicator). They must have done some research on this, perhaps while they are rolling this out. What I'd really love to see is LinkedIn opening up APIs, so that these verified profile identities could be shared using digital "Verifiable Credentials", as well as getting a verification tick based on other portable digital credentials - instead of scanning a passport being the only option. (I do understand that they need to bootstrap the ecosystem early on with such physical -> digital conversions.)

I can't immediately say that any specific technology would have prevented this since prompt hacking is such a rapidly evolving discipline. However, it also underscores the need for a permissioning and delegation-of-authority mechanism for AI agents as they take more autonomous control of doing actions on behalf of people. These agents are going to need identities as well. In a sense, it's good that this was discovered as part of a bug/hacking bounty. (Also Freysa isn't @fraser_again)

At the 40th @idworkshop, I walked away with one overwhelming thought: we are not ready for the Era of Hyperpersonalisation. The first Internet Identity Workshop (IIW) was launched in 2005. Now, 20 years later —celebrating the 40th IIW — we stand at the beginning of a generational fight for the future of our personal data. With IIW marking its 40th session, it's clear: we are entering an era where our digital identities are about to be redefined by artificial intelligence. Imagine a future where autonomous AI agents act on your behalf—filtering, negotiating, and even shaping your digital presence. I call this vision "autonomous self identity". These agents will not only model our hopes, dreams, and desires but also serve as dynamic representations of who we are in the digital realm. Yet, without robust privacy-preserving foundations, these digital twins risk being confined to "digital jails". Today, social media and ad networks continually test variations of ads or posts on us, using rapid-fire simulations to predict our behavior. Extrapolate that to AI agents conducting years' worth of psychological profiling in seconds. The potential for misuse is enormous, and the implications for personal freedom are profound. In March 2024, I wrote an internal memo forecasting three critical needs as AI agents grow increasingly autonomous (we ultimately made this vision public under an umbrella term called Verifiable AI or vAI): - Know Your Agent (KYA): Just as we have KYC (Know Your Customer) and KYB (Know Your Business), we need robust mechanisms to verify and credential AI agents. - Granular Permissioning: Current technologies lack the detail required to manage the permissions these agents will need. - Secure AI Wallets: Beyond facilitating payments, these wallets must store identity credentials and verifiable authorizations for both humans and AI. The most viable way this would happen is if AI apps or “agents” had wallets — not only to store secure means of making payments, but also for storing identity credentials from the humans they were working on behalf of, and the humans (or AI!) that had created them. What I didn’t know back when I wrote that memo is how this would ever work without an industry standard for how AI apps interfaced with the “outside” world. In the past four months, the centre of gravity has significantly shifted as advancements such as Model Context Protocol (MCP) and A2A (Agent-to-Agent Protocol) slot in the final pieces of the puzzle for making “autonomous self identity” possible. Despite these breakthroughs, at IIW 38 only one session touched on AI and identity; at IIW 39 there were only three. A complete redefinition of how our personal data and identity are handled online is not a distant threat! At IIW 40, maybe only 10-20% of some of the finest digital identity experts in the world had even heard about MCP. There were perhaps fewer than 10 sessions about how to tackle the new challenges that the onset of autonomous self identity would bring (three of those were run by me!). I don’t think our industry has gotten the memo on how crucial it will be to create a practical path forward in the oncoming generation of autonomous self identity apps. This new generation of AI-powered apps will be incredibly addictive and simple to use, and much like the way Facebook, Instagram, and TikTok snuck into our lives. If we don’t do something about it, they'll be extremely hungry for our personal data. And so it worries me that the kinds of solutions and thinking I saw at IIW 40 are not really fathoming how quickly the world is changing, and how privacy-first approaches CANNOT be subpar user experiences. We cannot seriously be telling people to scan three different QR codes or to care about the minutiae of different DID methods and protocols. It's probably fair to say that the companies that defined the Era of Big Tech and Social Media — Facebook, Twitter et al (all founded around the time IIW began) — won the battle on what "normal" feels like for privacy and confidentiality of our personal data. But last time around this fight happened, we didn’t have the plethora of privacy-preserving technologies that we now have at our disposal. It would take another 4 years after that first IIW for Bitcoin to be launched, and 10 years before a programmable blockchain like Ethereum was realised. The reason I bring those two milestones up is that the technological advances that enabled those two events to happen also sparked fundamental advances in cryptography that have brought so many privacy-preserving technologies to life, such as zero-knowledge proofs, selective disclosure, fully homomorphic encryption (FHE), BBS+ signatures, zk-SNARKs (like PLONK) and zk-STARKs. What if the next generation of digital interactions is not mediated by faceless algorithms, but by AI agents that truly understand — and protect — our identities? Can we envision a future where every digital footprint is secured by cryptographic assurances, and every interaction is governed by user consent rather than corporate convenience? Will we continue to let our privacy be shaped by legacy systems that encourage data exploitation, or will we embrace a paradigm where personal control and trust are built into the very fabric of the digital economy? The choices we make now will determine the kind of digital society we build for tomorrow. We have witnessed transformative breakthroughs in cryptography and digital identity over the past two decades. With AI now accelerating at an unprecedented pace, every builder in the privacy and identity space must act immediately. This is not about any specific company. This is not about any specific project or initiative. This is not about any specific technology or protocol. It’s about ensuring that the new era of autonomous self identity is built on robust, user-centric, and privacy-preserving foundations. The Era of Hyperpersonalisation is upon us, and we have one chance to get it right.

Had such a long chat with @Gkas23 on the latest decentralised identity developments at JP Morgan’s Project EPIC (Enterprise Privacy & Identity Composability) / Kinexsys and vice versa. Last time we met was @idworkshop in 2022/2023?! I love the spirit of experimentation and fast following to product that the JP Morgan team has always had in this space, and being deep into the research too.

Trust is a scarce commodity, and this framework is a good articulation of why my hypothesis is that systems for building and distributing “trusted data” will be the core of the technology industry in the next decade and beyond, especially as we deal with and use a lot more generative AI.

All of the sessions at @idworkshop 39. Collaborating with other @DecentralizedID contributors on initiatives like DIF Labs, standardisation for DID methods (and what sets particular ones apart). #iiw39

Most of our team are flying out today, so we unfortunately can’t be around for demos/judging at @HackmosHQ. I did have a panicked submission sprint at İstanbul Airport with 15 minutes on the clock for submission deadline AND boarding gate close 🏃🏽‍♂️🛫🥲 Good luck to all the hackers, and we looked forward to building this for real at @cheqd_io for a native decentralised identity layer for @cosmos & @cometbft 💪🏽

🫡 This needs to become a limited edition #IIW t-shirt ft. @csuwildcat @TBD54566975

I've got some personal news: I join today as CTO and co-founder at @VerimID, working with @fraser_again. We are a startup working on making self-sovereign identity ecosystems economically viable and rewarding for all actors involved in an SSI ecosystem. Read on for why 👇🏾

Always love seeing @cheqd_io DIDs in the wild! From @dock_labs earlier today, on their work on local-first, privacy-preserving biometric binding to digital credentials.

😎 Try out the @cheqd_io demo yourself, get a Verifiable Credential for #IIW 34, and read more about the background for how we built this... nitter.app/ankurb/status/15194148… 💡 Our starting principle for this #IIW demo: we wanted to build a non-custodial wallet, which works in a browser.

Superintelligence Summit by @ASI_Alliance with speakers from @oceanprotocol, @SingularityNET and other is absolutely PACKED. It’s only the start of the day, but a lot of discussion already on how the quality of datasets (among other things) impacts how this path to Singularity plays out 🤓

Great to meet the @LitProtocol team in person at #Devcon2024! One of the smartest teams we’ve worked with, and I really haven’t seen *any* solid competitors to what they do.

What on earth IS @cheqd_io? Our vision is to improve #SelfSovereignIdentity that @fraser_again and I have found lacking - and building first-ever solutions for them. Kinda like this meme describes us... 😂☠️ Come find us @IdentityWeek_ID: we're near Track 3 stage #IdentityWeek

Absolutely loved the energy, meeting people across identity/blockchain (including people from Cosmos, Cardano, Polkadot), and learning so much across the sessions at @EFDevcon. I think Bangkok was great in terms of access for people across the world - and not just from the US and Europe. It was also great to hang out for extended periods with some of the @cheqd_io team like our head of engineering @EEAnder1, as it’s so much easier to brainstorm in person. Lots of ideas came out of that! And more broadly to be collaborating with so many contributors to @w3c and @DecentralizedID specifications. I appreciated how much thought had clearly gone into the human/service design of holding a conference for 12,000 people: well-signposted volunteers, snack stations everywhere, diverse types of meeting and working spaces, an app with a super useful AI chatbot to throw questions to, well-signposted volunteers, translation guides (nice to know this was even a problem to be tackled - shows how diverse the audience was!) While it was said in a different context, I think it highlights what @0xstark said at the opening ceremony about “Ethereum handles the hardest problems first because it has gone the longest”. That applies particularly in the case of putting together an event like this one. (Wore my @StargazeZone T-shirt today and SO MANY people came to tell me they loved it…and I told them about @badkidsart.)

Useful numbers around the concept of AI model collapse from @RuffTimo: I was always curious about at what point does a model become garbage once you introduce AI-generated/synthetic content; didn't think the number would be as low as 7%! Raises an interesting question on *how* non-generated/non-synthetic data would be tagged. linkedin.com/feed/update/urn…

Consulting as a career is basically being a corporate version of TaskRabbit, where companies can hire exactly the same people they would internally but without putting them under their own payroll / headcount (so it’s easier to hire and fire them)

Many thanks to the @PrivacyScaling team for facilitating this round table discussion on decentralised/zk identity with some of the smartest people across the ecosystem, including fellow @DecentralizedID contributors. @VitalikButerin dropped by at this d/acc day breakout to say we need to avoid a situation where “everyone needs to KYC for everything”…which is indeed a risk with digital ID systems! Nobody wants to end up in a situation where a drivers license or a passport is needed to just browse the web. So it’s important to work on the systems that give users control, and leverage zk/selective disclosure technology.

⚠️ We have a security release for @cheqd_io out. All operators are recommended to upgrade to v1.4.2 of cheqd-node software. github.com/cheqd/cheqd-node/… This release is a security update to remediate the "Huckleberry" vulnerability in ibc-go. Please see our security advisory notice for further details. This is a non-state breaking release and does not require a coordinated upgrade. Node operators are suggested to update their nodes regardless to remediate the vulnerability. You can use the interactive installer to carry this out, or by manually replacing the cheqd-noded binary wherever it is on your system.

Sessions on Day 2 of #iiw39 @idworkshop. Later today with @kimdhamilton, @AnonyomeLabs and others: rapidly-emerging use cases of decentralised identity in “Verifiable AI”

Running a session on AI working group & efforts at @DecentralizedID later today with @kimdhamilton and Andor. Also loved the quick demo from @peacekeeper on did:scid 🤝 did:cheqd compatibility.

All sessions happening at @idworkshop end today, in case it helps people figure out where to go. #IIW

Data for 33 million French residents stolen due to a breach at two large health insurance providers. 😦 If it’s true that the hackers phished healthcare professionals’ and used their login details to scrape *33 mn* records, this is an incredibly bad security architecture. You would hope that any company that keeps such sensitive data to have measures in place to detect when a threat actor is downloading and exfiltrating gigabytes of data. This is why we cannot trust current data controllers, and need to have systems where people control their own data vault. euronews.com/next/2024/02/08…

Last day of #iiw39 @idworkshop - I got a picture of the sessions although they seem to be a bit in flux. Running a technical / brainstorming workshop with @peacekeeper and @dbluhm_ on supporting more DID methods in ACA-Py.

It's been a HUGE quarter for the @cheqd team: We’ve upgraded our network’s identity features, bridged to Ethereum, plus much more… 🚀 blog.cheqd.io/q1-round-up-fr…

Spotted @rosspower in Prague 👀 (As it happens, “Rossman” is exactly what he’s called when he visits kebab shops.)

Special edition book of d/acc book from @VitalikButerin if you attend the d/acc discovery day on Level 2 #Devcon2024

“You need to be super careful before even considering writing anything on chain, since you can’t unwind that” - @Tom_Viera (important to drill this message in, since this is often the mistake a lot of web3 apps make) Session moderated by @provenauthority on making Web3 more intuitive for mass adoption.

I really enjoyed having this conversation with @HouseofChimera, especially around how important proof-of-human generated content would become. 64 countries around the world have some form of national elections this year, including the US, UK, and India. I fear that at least one of these elections will get swayed by AI-generated content, which will make content credentials to denote what %age of a text/image/video is human-generated vs AI-generated a lot more important in the discussion.

Thoughts from Day 1 & 2 of @WebOfTrustInfo Rebooting Web of Trust #RWOT... 1. I was surprised by how little generative AI was a topic of conversation, definitely an outlier from the other conferences I've been to this year. The sense I get is that it's still seen as a shiny toy that makes coding and writing easier, rather than something that digital credentials could be applied to. (There's some really cool stuff that @C2PA_org is doing in this space around content provenance and authenticity, which can survive editing/cropping for legitimate purposes.) 2. Japan started looking at SSI because earthquakes/tsunamis are an existential threat to business continuity of data/data centres. Hence, giving people a copy of their data was seen as a way of leapfrogging how data is handled and managed. Still early days of course, but a cool insight on such a different driver for adoption compared to Europe, where it's driven more by increasing competition in the market / interoperability. 3. There were some interesting conversations around how to understand whether an "expert" or an "authority" is actually an expert/authoritative, without falling back to centralised mechanisms. (This is a topic I touched on in my last year's advanced reading submission for dynamic and decentralised reputation scoring: github.com/WebOfTrustInfo/rw…) 4. A lot more focus on sustainable technology, not just ethical/privacy-preserving technology this year. I get the sense that the more obvious utility of generative AI technology (as well as climate change) has made the conversation more urgent and tangible. With blockchain, either you fell into the #Web3 camp where decentralisation was "obviously good", or thought the energy footprint of it was "obviously insane" because people questioned its utility. It's a much harder trade-off on tech progress vs climate when more people get why generative AI is useful. 5. The definition of what "good" or "bad" varies a lot. The example of Worldcoin was brought up, where this group of people think it's obviously bad. On the flip side, Worldcoin has users (even if acquired by questionable means) and on a completely different axis of trying to establish universal basic income, could be considered "good" - even if "bad" from a privacy perspective. 6. There's a lot of emphasis on direct/peer-to-peer connections. I do wonder though about intermediaries sometimes being valuable, e.g., eBay or Airbnb being intermediaries in ecommerce/rental housing (where there's someone to turn to and complain when things go wrong). An ideal system would be (more) direct and perhaps have some of the nicer attributes of intermediaries (someone/some mechanism of complaining and rectifying mistakes). 7. We did a past/present/future idea board with stickies. This was super fun to capture the mood of the attendees: - 1990s: People mostly spoke about new tech, like email, sites, instant messengers etc. Very nostalgic. - 2000s: Post 9/11 made surveillance capitalism normalised. Web-first started becoming a thing. - 2010s: The slow rise of Big Tech. - 2020s: A lot of despair, not just in the tech sense but a general atmosphere of sociopolitical instability. A sense that people are no longer just talking about the tech, but also talking about the impact of tech on society and vice versa. - 2030s+: A hope that there's a ray of sunshine after the era of Big Tech, where people are more in control of their data and privacy is a human right.

Hiring a new role for my team @zama_fhe, hmu if you know/are a good SRE/DevOps engineer. In particular, I'm looking for people with experience of running with managed infra APIs. Does not need to be Paris-based, but would prefer to have someone in +/-4 hours of UTC. Apply here 👉🏽 jobs.zama.ai/companies/zama/…

Since it’s cheaper to distribute via pharmacies, it’s also the default path they nudge people towards when trying to order a home kit.

Loved getting a @badkidsart rendition of my @bitkidsart drawn as physical artwork at the @StargazeZone event! 😍 I got the Bit Kid because it reminded me of my birthday dressed as a dinosaur 🦖🥲 When @Toatse offered me the opportunity to own one, I knew I had to jump on getting it.

I could NOT have wished for a more capable and dynamic scrum master, then operations lead, and then senior product manager. Ross made his mark across so many different parts of @cheqd_io as a company, and I can't wait to cheer him on for his next on-chain or off-chain adventures. ❤️

QR codes are problematic because they are not easily human-verifiable that it’s taking you to the right party you expect. There’s prior art on this: - In China, where WeChat/Alipay are big for payments, scammers replace the actual QR code for a shop with their own. Until you reconcile payments, you might not even realise the switcheroo. techinasia.com/fake-qr-code-… - Many scams/hacks in crypto where QR codes on sites (encoding addresses to send payments to) get replaced with fake ones from scammers. I heard this story where at a crypto conference, someone created a QR code with caption “Don’t scan this QR code”. If you scanned it, it took you to a page that said “Don’t connect your wallet” as a caption beneath a wallet connect button. If you connected anyway, the app would drain your entire wallet…and then return all the funds after 5 mins. It was designed to make a point. This is the topic of my paper at RWOT this year, ie, what the human factors / ways in which scammers will try to compromise decentralised ID systems. 🤓 IMO, decentralised ID wallets will need an equivalent of the Google Safe Browsing list, which is one of the few open services where fraudulent domains get reported. (This and other lists are used by browsers when they show “Warning: this site might be fraudulent”) We’d need, as an ecosystem, a way of intercepting and warning users if they *think* they are connecting to their bank (via DID), but actually being taken to a pretender. This hasn’t been a problem so far because the DAU/MAU numbers aren’t high, but there’s a lot of lessons to be learnt from what fintech apps (like Alipay) or browser lists had to do.

While the formal Day 1 agenda for #iiw gets written up, here’s a quick look at the pre-lunch sessions @idworkshop (I’ll be doing the session on “Self-Sovereign Identity is highly-centralised: how do we fix the rotten core of reputation)

Many hot-takes in the past 24h of #Consensus2022 have been along the lines of "haha what is #web5 anyway is this a Fibonacci sequence 🫠" You can hate the game behind buzzword bingo, but THIS is why you should pay attention to @blocks' announcement 🧵👇🏽 nitter.app/TBD54566975

This is an exciting day for @C2PA_org Content Credentials, with a prominent phone manufacturer coming onboard to support the standard. Content Credentials, if you're unfamiliar, are like a "chain-of-custody" for what edits happen to an image, for example, how much of it is AI vs human-generated. Time for @Android and @Apple to come onboard, to natively support these at the OS level.

I didn’t even *know* this was a thing but @Spotify has a feature which shows the most shared podcasts…and @provenauthority’s podcast on @bankless with @VitalikButerin is one of THE most shared podcast episodes!!! 💃🪩🤯 You should listen if you haven’t open.spotify.com/episode/0lS…

Excited to be giving a pre-@WebOfTrustInfo talk along with @TimoGlastra of @AnimoSolutions on how we’re working together to support the AnonCreds digital identity credential format on @cheqd_io using @Hyperledger Aries Framework JavaScript. Sign up here: eventbrite.co.uk/e/showcasin…

This paper co-authored by members of @DecentralizedID, @trustoverip, and various members of the @w3c Credentials Community Group was a great example of how many of us in the tech industry are collaborating for new, interoperable forms of digital identity. The EU Digital Identity Wallet initiative which aims to have reusable "Digital Identity for all Europeans". Specifically, in this paper/response we addressed a few common concerns and misconceptions that people tend to have about decentralised identity.

I've noticed a funny trend: a lot of job applicants now write answers FOR ChatGPT…assuming that's who's reviewing them. All the hiring on my team at @zama_fhe is 100% organic 🌱 No AI CV filters, no GPT-powered recruiters...just a real, slightly overworked human (me) reading every application 🙋🏾‍♂️😅 If you want to bring an entirely new generation of FHE-powered privacy-preserving apps to life, we should talk. 👉🏾 jobs.zama.ai

Totally knocked out by a cold and recovering in the south of France (the pharmacist said “Ah yes, I’ve heard of this from the English” when asked for Lemsip) and I’ve been reading “Why the Dutch are Different” by @bencoates1. I chortled hard at the Belgians thinking of the Dutch as “Tintin-reading simpletons” (given that Tintin is Belgian). 😅

Love the pirate-like @Ledger flag on the rooftop, and the documentary about what it took to manufacture Ledger Stax (would’ve love to know even more how the innovations on linking up the secure element to e-ink touchscreen were done).

Kicking off Day 1 with a lot of talk on AI @idworkshop…and the cool thing is organically discovering new developers building on @cheqd_io