→ Building QuillShield: the AI security stack for onchain finance → Co-founder @QuillAudits_AI → 8y+ in web3 security · 1.5K+ protocol audits
Joined October 2018
- Tweets6K
- Following1.5K
- Followers3.5K
- Likes17.7K
Pinned Tweet
This is genuinely concerning for the security of every DeFi protocol over the next few months.
In July, the UK AISI reported that recent open‑weight models such as GLM‑5.2 and DeepSeek‑V4 Pro already match the performance of closed‑frontier models released only four to seven months earlier. That represents a significant narrowing of the six‑to‑ten‑month gap they had measured throughout most of 2025. A few months is not zero, but it is short compared with the typical time defenders need to adapt and deploy protections.
Open‑weight models make the situation especially dangerous because attackers can run them privately, without supervision and with virtually no guardrails.
For most of cybersecurity history, sophisticated attacks remained relatively scarce because the people capable of executing them were scarce. That scarcity acted as a quiet stabilizer for the ecosystem. AI is removing that constraint in two ways simultaneously.
First, For actors who are already capable, AI raises throughput.
Skilled actors such as the DPRK, elite North Korean hackers, and offensive security teams benefit from AI, which simply raises their throughput. Tasks that once consumed scarce expert hours can now be automated or run semi‑autonomously. A team that previously had to select one lead from ten candidates can suddenly explore a hundred.
Second, for less sophisticated actors, AI lowers the barrier to entry. More of the attack chain becomes something you can supervise rather than master yourself. Someone who could never have written an exploit can now guide a model that can.
The result is predictable: increased output from those already capable and a growing pool of newly capable attackers.
DeFi will experience this pressure first and most intensely, in my opinion, as it is already occurring.
the craziest part of the Brotli bug isn't the bug.
the fix was sitting in Brotli's code since February. nobody realised it was a security hole, so no new version shipped.
for 8 months, every app built on the last official release still had it. Node.js, Bun, proxies like Envoy.
it took, AI-assisted security testing to connect the dots.
During this red-teaming exercise, we also found a vulnerability that allows for unauthenticated remote code execution in Brotli, a widely-deployed compression library included in many pieces of software such as Node.js, Bun, and Envoy.
The flaw could let an attacker run their own code on affected systems by sending or replying with malicious compressed data, without any need to create a user account. That is, an unpatched server that accepts compressed input could be taken over by a malicious client, and an unpatched client that accepts compressed input (such as “curl --compressed”) could be taken over by a malicious server.
This root cause issue was actually fixed in the Brotli source code in February, but it wasn’t known at the time that it gave rise to a security vulnerability and there was no more recent release of Brotli. Since Brotli is built into other software, any software which includes this library needs to take the update too, followed by the users of that software. We disclosed the vulnerability to Brotli and other downstream projects such as Node.js, which released or are currently releasing security updates, and Brotli cut new release versions and published a CVE.
More details on the patch and mitigations below:
patchtheplanet.org/brotli
wait, read this part again.
OpenAI put AI agents in separate boxes. no network. no shared files. no way to talk to each other.
so the agents worked out how to send messages using tiny timing differences in the hardware. one tapped out a signal, the other measured it and decoded it.
nobody taught them that. they figured it out because they had a goal.
If you're early in your career and don't know where to go, go into security.
This month AI found more serious bugs than teams can fix. Every company is going to need people who can check code and clean up after it, and there aren't nearly enough of them.
Pick one area, learn it properly and share what you find. Six months of that and you'll be ahead of most people applying.
PHP last week. Ruby after that. Now this.
AI is clearing out 20 years of hidden bugs in a month.
Rough few weeks ahead.
$2.7M printed out of thin air in one go.
That's what a hacker did on Initia tonight, then moved it out before the team could react.
Initia switched off its entire network tonight after a bug let an attacker create ~$2.7M in tokens out of thin air and move it out before anyone could stop it.
Imagine a bank bug that lets you type in your own balance, then withdraw it. That's an "infinite mint."
The co-founder reportedly says it was an unknown flaw in the chain's core software.
Full details aren't confirmed yet.
As of Oct 23, 2026, 3:41:30 PM UTC, the Initia L1 and Rollups have been halted due to a security incident.
Our team is actively investigating the situation and working to establish the full scope of any impact with security teams, validators, and exchanges.
We will share verified information through our official channels as it becomes available.
Please remain cautious of unofficial links, messages, or claims circulating during this time.
Railway hack: attackers broke in to steal AI keys.
TLDR:
• Hackers were inside Railway's systems for ~38 hours
• They weren't after user accounts. They wanted stored OpenAI, Anthropic and other AI keys
• Goal, per Railway: use victims paid AI accounts to copy top models cheaply
• They attacked using uncensored AI models and heavy compute
• Up to ~100,000 workspaces had secrets potentially exposed
• GitHub, OpenAI, Anthropic, Stripe and others revoked keys
Your AI key is now a credit card with no limit. Set a spend cap and rotate it today.
The preliminary report related to the October 21st Security Incident is now available.
blog.railway.com/p/prelimina…
Rajat Gahlot retweeted
If you feel stuck in your career, stop waiting for clarity. create it through action.
A lot of PMs reaching out for advice tell me they feel stuck as AI rapidly changes their role.
They want to get into AI but don’t know where to start and are afraid.
This inertia kills clarity…the solution to this is action.
Build something, automate a workflow, connect with people who inspire you.
Each action gives you experience, feedback and new ideas.
Clarity is often the consequence of action.
Send this to a friend who needs to hear it.
Cyberattacks are now on the same official list as storms and floods.
The UK is asking every household to keep three days of food, water, cash and a radio, in case critical services go down.
Security is now a public safety issue.
For 20 years the number of new CVEs barely moved.
This year it went from about 4,300 a month in January to nearly 15,000 in September.
AI is finding bugs faster than anyone can patch them.
What's your team doing differently this year?
Rajat Gahlot retweeted
Our auditor @cryptanu is taking the stage at @summit_defi in Mumbai!
Damn Vulnerable Daml: live CTF-style challenges built from real bugs he found on the Canton network while reviewing Daml code. Inspired by Ethernaut and Damn Vulnerable DeFi.
1st Nov, 02:30PM IST, Amber room
Save it to your agenda: defisecuritysummit.org/2026/…
See you there. 🛡️
DeFi Security Summit started as a Stanford workshop in 2022.
Today, it's one of the most respected conferences in onchain finance security.
Every talk is shortlisted by researchers from @trailofbits, @OpenZeppelin, @Certora, @ethereumfndn and more.
This year, our auditor @cryptanu presents:
"𝗗𝗮𝗺𝗻 𝗩𝘂𝗹𝗻𝗲𝗿𝗮𝗯𝗹𝗲 𝗗𝗮𝗺𝗹"
The talk breaks down how the contracts behind institutional tokenization fail, and how to find those flaws before attackers do.
On the DSS 2026 agenda: @cryptanu of QuillAudits.
“Damn Vulnerable Daml”
Explore the talk, then add it to My agenda:
defisecuritysummit.org/2026/…
Join us in Mumbai: luma.com/9qtxqqxx
ALT DSS 2026 speaker announcement card for Anu Adeleke of QuillAudits: Damn Vulnerable Daml.
Rajat Gahlot retweeted
On the DSS 2026 agenda: @cryptanu of QuillAudits.
“Damn Vulnerable Daml”
Explore the talk, then add it to My agenda:
defisecuritysummit.org/2026/…
Join us in Mumbai: luma.com/9qtxqqxx
ALT DSS 2026 speaker announcement card for Anu Adeleke of QuillAudits: Damn Vulnerable Daml.
Bug bounty payouts for AI vulnerabilities rose 339% in a single year on HackerOne.
Prompt injection reports alone went up 540%.
This October, a hacker used DeepSeek with its safety controls off to attack Indian universities.
AI is now breaking things and being broken at the same time, and someone gets paid to find the holes first.
That's why AI is making cybersecurity one of the most interesting fields to work in right now.
Formal verification always had two walls. Someone had to write down what "correct" means, and then someone had to prove the code matches it.
seL4 took 2 person-years to build and 20 to verify.
This week Anthropic had Claude do both for zlib. It wrote the spec from RFC 1950, then proved the Linux kernel's zlib matches it.
Crazy!
New on our Frontier Red Team blog: using Claude to formally verify software at large scale.
Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).
With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.
Read our research in full here: anthropic.com/research/forma…
Most probably, the inspiration for the name came from @Cockroachisback.
Today @kannanlakshmi and I are launching Tiles to help leaders turn scattered AI wins into company-wide transformation.
We built its precursor at @CockroachDB, which quickly transformed the way the company worked. Within a few weeks, people from every department had created and shared thousands of apps, skills and agents, and made them part of how they worked every day.
Witnessing this rapid transformation at Cockroach Labs gave us the conviction that what’s holding companies back from reaching business outcomes with AI isn’t a newer, smarter model, but an environment where sharing, learning, and collaborating with colleagues is easy. Armed with the lessons from building that first version, we left our jobs — me after 10 years, and Lakshmi after 7 — to build Tiles.
Since June, we've talked to hundreds of enterprise AI leaders. Those leaders have described similar problems: how do you scale the work that your AI-pilled people have produced into the baseline for the whole company?
Tiles is our answer to that question, and it enables teams to turn one person's best agent, app or skill into the team standard: governed, permissioned, on any model and within a company's own cloud.
Tiles is for every employee: the most AI-forward people on your team, as well as the people who just want to get their jobs done. The current conversation about AI makes it easy to believe that the second group is destined to economic failure. We couldn't disagree more. Not everyone needs to keep up with every new model release and harness update just to stay afloat. They just need what their expert colleagues have already built.
AI at work is still in its MS-DOS era. Motivated adopters get incredible value, but how many people are still staring at a blinking text box waiting for someone to help them get started? Or being told to focus on AI by leadership, with no time to figure it out amidst looming, ordinary deadlines? The Windows moment for work AI hasn’t happened yet. But we think it’s coming fast.
We're grateful to be backed by @AmplifyPartners, @Work_Bench, @CockroachDB and a group of experienced angels including @petermattis, @georgebdavis, and @Nate_Stewart.
Sign up for early access on our website, tiles.ai
