←
Matt Hardy692 posts
Matt Hardy avatar

Matt Hardy

@mdahardy

cto @proofofhuman

◷ Joined April 2014▧ 159 Photos and videos
Matt Hardy avatar
Matt Hardy@mdahardy·3h

It's funny how path dependent company decisions can be! Many years ago when I was first learning React, the Youtube series I was following used Netlify to deploy their example site. Since then I've defaulted to using Netlify for all my personal projects, including the first iterations of Proof of Human. Netlify has continued to improve as we've grown, and we're happy to still be a customer!

we're proud Netlify users @proofofhuman cc @mdahardy

Matt Hardy avatar
Matt Hardy@mdahardy·5h

Post attachment

So many times I have had this interaction with normies where they don't understand what the jackfriks I'm working on as I'm just making random things for fun and messing around with things And to them whatever I'm doing looks like a complete and utter waste of time, it has no function, no direct rewards, it's not making money, it's not successful But so many times those projects evolve or lead to things where there is a massive payout or success after But most people don't have the ability to see that far ahead, and they're risk averse so even if you tell them "I don't know maybe this becomes something, or maybe not, it doesn't really matter" that's simply too uncertain enough for them to accept. Which is of course why they love employment so much, it's stable and certain. The problem with normies is that they will quickly shut this creative play down in themselves and even worse in others. And that way whatever that thing is, it won't ever grow into anything because they killed it before it ever got a chance, proving their point Almost all my projects started by just messing around with no real goal, just playing for fun But it requires some kind of peacefulness because the moment you start putting a requirement of success on to something new, the thing instantly dies Creativity is really magical, and it messes with you if you try control it or expect things from it

Elisa Kreiss avatar
⟳ Matt Hardy retweeted
Elisa Kreiss@ElisaKreiss·Nov 10

I'm recruiting a PhD student to join me at UCLA Communication & the UCLA NLP group! I’m especially (but not exclusively) looking for students interested in how models learn from text and image data, and how that learning gives rise to the behaviors we see in trained models. (1/4)

Fernanda Palacios avatar
⟳ Matt Hardy retweeted
Fernanda Palacios@fernandaplcs·Nov 9

love this introduction on density plots from @mdahardy, thorough and intuitive. it also shows the power of visualization to understand complex data

Post attachment
Mayank Agrawal avatar
⟳ Matt Hardy retweeted
Mayank Agrawal@mayankagrawal·Nov 5

A competitor bot-detection company wrote a self-proclaimed hit piece on a recent @proofofhuman benchmark. The criticism and condescension was based on fabricated quotes, while the article actually agrees with all our real quotes and conclusions. Grateful to work with people like @milenamr7 who demonstrate poise, class, and rigor. Disappointing to see fake quotes, but it just highlights the market opportunity in bot detection and how customers need better people to work with.

We appreciate the read and critique, and we believe we are in agreement that a technical benchmark is very useful, and is not a full comprehensive product evaluation across all anti-bot use cases nor all forms of adversarial traffic. But we are concerned by your misrepresentation of our report, including direct misquotes and over-generalizations. The main thesis of the criticism is summarized in the following quote from your blog post:"What the benchmark really tells us is how some publicly accessible configurations reacted to a small number of isolated automated sessions on a new website with almost no historical traffic. That's a valid experiment. It's just a much narrower conclusion than "we detect 99% of bots."" We agree. Our benchmark is an evaluation of five systems in a controlled experimental setting, and our article consistently states that. We never claim that "we detect 99% of bots" or definitively solve "which anti-bot product is best". Both misquotes come from your post, not our report. On false positives - we agree this matters, and we acknowledge it in our limitations. In this report we look specifically at failure modes of agentic detection, which many of these systems advertise. Including humans does not change which systems miss which agents. On benchmarks being narrow - we also agree that a full, comprehensive test is always ideal, but evaluations are generally done on a subset of specific, controlled conditions. Our work used a paradigm that was accessible to an Internet-native user: AI agents completing a survey, and we found clear weaknesses in widely used systems that are worth reporting. We are happy to engage further and address all your points in detail in a lengthier format if that would be of interest.

Matt Hardy avatar
Matt Hardy@mdahardy·Nov 4

Unfortunately reviewer 2 follows you for life

We appreciate the read and critique, and we believe we are in agreement that a technical benchmark is very useful, and is not a full comprehensive product evaluation across all anti-bot use cases nor all forms of adversarial traffic. But we are concerned by your misrepresentation of our report, including direct misquotes and over-generalizations. The main thesis of the criticism is summarized in the following quote from your blog post:"What the benchmark really tells us is how some publicly accessible configurations reacted to a small number of isolated automated sessions on a new website with almost no historical traffic. That's a valid experiment. It's just a much narrower conclusion than "we detect 99% of bots."" We agree. Our benchmark is an evaluation of five systems in a controlled experimental setting, and our article consistently states that. We never claim that "we detect 99% of bots" or definitively solve "which anti-bot product is best". Both misquotes come from your post, not our report. On false positives - we agree this matters, and we acknowledge it in our limitations. In this report we look specifically at failure modes of agentic detection, which many of these systems advertise. Including humans does not change which systems miss which agents. On benchmarks being narrow - we also agree that a full, comprehensive test is always ideal, but evaluations are generally done on a subset of specific, controlled conditions. Our work used a paradigm that was accessible to an Internet-native user: AI agents completing a survey, and we found clear weaknesses in widely used systems that are worth reporting. We are happy to engage further and address all your points in detail in a lengthier format if that would be of interest.

Milena Rmus avatar
⟳ Matt Hardy retweeted
Milena Rmus@milenamr7·Nov 4

We appreciate the read and critique, and we believe we are in agreement that a technical benchmark is very useful, and is not a full comprehensive product evaluation across all anti-bot use cases nor all forms of adversarial traffic. But we are concerned by your misrepresentation of our report, including direct misquotes and over-generalizations. The main thesis of the criticism is summarized in the following quote from your blog post:"What the benchmark really tells us is how some publicly accessible configurations reacted to a small number of isolated automated sessions on a new website with almost no historical traffic. That's a valid experiment. It's just a much narrower conclusion than "we detect 99% of bots."" We agree. Our benchmark is an evaluation of five systems in a controlled experimental setting, and our article consistently states that. We never claim that "we detect 99% of bots" or definitively solve "which anti-bot product is best". Both misquotes come from your post, not our report. On false positives - we agree this matters, and we acknowledge it in our limitations. In this report we look specifically at failure modes of agentic detection, which many of these systems advertise. Including humans does not change which systems miss which agents. On benchmarks being narrow - we also agree that a full, comprehensive test is always ideal, but evaluations are generally done on a subset of specific, controlled conditions. Our work used a paradigm that was accessible to an Internet-native user: AI agents completing a survey, and we found clear weaknesses in widely used systems that are worth reporting. We are happy to engage further and address all your points in detail in a lengthier format if that would be of interest.

Apparently it’s not only AI labs that are good at gaming benchmarks. The anti-bot industry can do it too. I wrote about what bot detection benchmarks actually measure, what they miss, and what I’d look at instead. blog.castle.io/how-to-win-yo…

Post attachment
Jesse Spafford avatar
⟳ Matt Hardy retweeted
Jesse Spafford@jessespafford·Nov 2

Incredible, Elon trotted out the stolen joke irl to impress Jensen Huang:

Post attachment

This is crazy but Elon Musk just stole my friend's joke and even emoji reacted to the original showing that he had read it

Post attachmentPost attachment
Matt Hardy avatar
Matt Hardy@mdahardy·Nov 3

This joke is my life's greatest work. X has hidden from my profile and from search though 🤷

Incredible, Elon trotted out the stolen joke irl to impress Jensen Huang:

Post attachment
Proof of Human avatar
⟳ Matt Hardy retweeted
Proof of Human@proofofhuman·Nov 2

Proof of Human is coming to @UCLA this Friday, November 6th! Our co-founder and CEO, @mayankagrawal will present our framework for distinguishing humans from AI through behavioral dynamics, including patterns in perception, control, and decision-making. Drawing on process-tracing data from cognitive tasks, the talk will explore how behavioral signals can reveal computational signatures unique to human cognition and establish a more robust foundation for proof of personhood.

Post attachment
Fernanda Palacios avatar
⟳ Matt Hardy retweeted
Fernanda Palacios@fernandaplcs·Oct 23

banks should be the first to use @proofofhuman. we're not ready for agents to transact on their own

SCOOP: Major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts, including some attempting to take actions that appear to be independent of user instructions, sources told Axios. The findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond, multiple industry and bank executives told Axios. Several banks are now preparing for worst-case and even “day after” scenarios, including how to rapidly secure or restore customer accounts if things go wrong, the people said. Several bank executives told Axios they're struggling to distinguish between human users and AI agents. One executive said banks also struggle to determine whether agents are authorized by customers or potentially operating without their permission, more akin to rogue agent behavior. Read my latest for Axios here: axios.com/2026/10/22/ai-agen…

Matt Hardy avatar
Matt Hardy@mdahardy·Oct 23

When we started working on @proofofhuman last year human verification was still a relatively niche problem. Things are changing fast!

SCOOP: Major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts, including some attempting to take actions that appear to be independent of user instructions, sources told Axios. The findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond, multiple industry and bank executives told Axios. Several banks are now preparing for worst-case and even “day after” scenarios, including how to rapidly secure or restore customer accounts if things go wrong, the people said. Several bank executives told Axios they're struggling to distinguish between human users and AI agents. One executive said banks also struggle to determine whether agents are authorized by customers or potentially operating without their permission, more akin to rogue agent behavior. Read my latest for Axios here: axios.com/2026/10/22/ai-agen…

Proof of Human avatar
⟳ Matt Hardy retweeted
Proof of Human@proofofhuman·Oct 21

Our bot detection benchmark reached the front page of Hacker News (@hackernews) yesterday! 🚀 We tested five leading bot detection systems against modern AI agents, including CloakBrowser, Chromium, Grok, and Perplexity. Proof of Human accurately detected between 95–100% across all five adversaries, with most alternatives specifically struggling against AI-native browsers. We're excited to contribute open research to this problem -- and grateful to the Hacker News community for the discussion!

Post attachmentPost attachment