php devs had both mythos and daybreak red and they were still outrun by wiz calif using gpt-5.6-sol. you really can't buy the skill.
wait, PHP was actually scanned by Anthropic’s Claude Mythos earlier this year. they also had access to OpenAI’s Daybreak Red for validating exploits.
the PHP Foundation revealed this JUST yesterday.
keeping frontier cyber models in the hands of a selected few doesn’t seem to be working well for the internet.
thephp.foundation/blog/2026/…
Good reminder that bug bounties can completely change your life in a matter of months.
One bug can pay more than a year’s salary, and one great year can change your financial future entirely.
From homeless to $1.4M in 11 months. Few careers offer upside this fast.
copper screw retweeted
i would rather be an optimist and work hard than a pessimist posting about why things won't work.
it's much more difficult and the most likely path is failure, but society fails if people don't try.
no amount of "it will never work" essays will drive society forward.
1/ How I use AI in smart contract audits (2026)
From 2023–2026, I entered 57 contests, reached top-3 30 times, won 21 and earned $627k in prizes.
109 highs + 156 mediums. 53 of them solos.
86 private audits.
This came from reading every line and building the mental model.
copper screw retweeted
Can we stop using terms like top security firms, top auditors, or top whitehats to indicate how difficult a exploit was after they missed it?
> the vulnerability had evaded years of scrutiny by many of the world’s best cryptographers.
lol
copper screw retweeted
My take on the Zcash bug.
I never worked on Zcash as a target because I didn’t didn’t even know they had a BBP.
It it were a 2M bounty on Immunefi I think I would have been very likely to find it back in 2023.
I found three similar ZK bugs that year, plus 3 others in verifiers. All loss of TVL.
I read every single line of circuit code (multiple times) that had a an Immunefi BBP with max bounty >= 1M between 2023 and early 2024 and earned more than 1M from that effort, in just one year. Made even more later.
During all that time I didn’t even know there was a Zcash bug bounty. (Or Solana for the matter!)
Im sure there were other people doing this that found ZK bugs too.
I dont think anyone ever spent that kind of effort on their code base.
Most of those code bases I found bugs in had been heavily audited (that’s true in general for most findings).
“AI found this bug humans missed for 4 years” is a catchy headline
The reality is for 4 years they really didn’t incentivize humans properly.
And there are likely many other ZK bugs out there not found yet by humans, AI-assisted or not.
I wonder if any LLM could have found it without training of my own disclosures from 2023, ToB’s articles, RareSkills content etc
You can hope you will find all the bugs with an AI trained on data from 6 months ago at best — and hope your code only has bugs similar to publicly disclosed ones.
Or you can “risk” having to pay something like 1M instead of insta-crashing down 50% of market cap — or worse.
If exploited those are protocol-ending bugs.
copper screw retweeted
"Those who are cowardly in love are doomed to see the person they love with someone else."
There are so much money to be made in web3 security, it's insane
Study the craft. Many people did it - learn from them, follow their steps, or find your own path. Keep going, I am yet to see someone going all in into web3 security and not making it🫡
copper screw retweeted
Coming back to the “What’s the biggest thing bug bounties gave you?” Question,
Money aside, BB taught me the art of not giving up and the art of not giving a fuck. Hunting for bugs has been a spiritual journey more than anything else...
If you are in the beginning of your journey, you probably see other people making it big. Making big dollars like @Ehsan1579. You sit there and ask yourself just “How”. What do they have that I don’t? As rejections roll out and your reports get closed one after the other one, all you ever feel like is giving up.
You might feel defeated, unsatisfied, all the small chatter in your head is telling you to give up… But your EGO won’t let you. EGO isn’t all bad, as society likes to condemn it. I’d go as far as to say that ego is the key ingredient in an industry as cut-throat as BB.
In moments like these, your family, your friends and whoever you consider close will try to tell you that maybe it’s just not for you.
Solution? -> cut everyone. Don’t let anyone get to you and put your head down, analyze what is going wrong, how to get better and try new strategies. That’s the only difference between the ones who make it and the ones who don’t.
Do not take “no” for an answer. If a project closes your report but you know you are right, make sure you are right and then ask for mediation with all the facts you have. You’d be surprised how many projects close valid in-scope issues. Same thing applies if they try to reduce the severity or pay you “peanuts” for what it’s worth.
I’ve been working in DEFI for years now, learning the ropes, building projects. All of them were a fluke until I tried Immunefi. I worked every waking minute for 8 months straight without any payout.
Then… in September I got my first bug confirmed. A “Low” for $2000. I was ecstatic, I was excited. For literally 3 days, I was listening to music just staring and the “Confirmed” ticket.
You see bug bounty is a bit like the casino, the difference is instead of betting money, you bet time (sometimes it does cost money too) and the payouts hit like a truck and you never quite expect it fully.
It’s like a drug. And like a drug, you want more of it and you soon become accustomed to a certain level (“the tolerance”). They say you always chase the first high and it’s very true for bug bounty as well.
The months following I was finally averaging at least 1 paid report per month. That was until January. I’ll never forget it, I had 9 confirmed reports in a row. The “High” hit soooo strongly. But with every high comes a low. Doubts cripple in, you start rationalizing it. You tell yourself that “I just got lucky”. The bug Immunefi posted on their page was from January.
This cycle repeats endlessly. Turns out that February was an even better month for me. The thrill of the unknown and the unexpected is what makes it fun. I do not think I would be doing this if you took that out of the picture.
It was a journey, a very difficult one I’d might add. But in the end, it was worth it. It built me into a stronger more resilient person. It thought me patience.
Reports being closed hurt. Probably one of the worst types of rejection out there. But at some point in time you get used to it and have to learn how to detach yourself from the emotions and trust the process. It’s a numbers game after all.
I am waiting for the leaderboard to update now (long due)! I’m curious to see if I finally made it into the top 100 of whitehats of all times. (Currently sitting at $175k in earnings from 22 reports)
It’s all just a matter of time after all (;
copper screw retweeted
Trying to build the best tool to replace human auditors with AI is a race to the bottom.
You're spending all your capital getting good at catching bugs everyone will catch, progressively cheaper, with cost approaching zero fast, and you have no moat.
Get out while you can.
copper screw retweeted
the figures referenced in the post are entirely misleading. There was no impact realized from this issue. Zero user funds were affected and zero addresses were compromised.
My response: Are you suggesting I should have actually exploited the bug and caused real damage before coming to talk to you?
For the stated vulnerability to work in practice, it would require execution of several suspicious transactions that would have an extraordinarily limited impact.
My response: You should know better than anyone that on a Cosmos-based chain, a single transaction can pack multiple messages. Just one transaction is more than enough to completely drain multiple whale accounts.
Injective has dynamic rate limiting functionalities which are applied automatically based on our live monitoring systems. This functionality has been live on mainnet since last year and is publicly available in our code base.
My response: First, this has nothing to do with the vulnerability itself. Rate limiting doesn't stop attackers from stealing funds. It only slows them down when they try to bridge those funds over to Ethereum. Second, when I submitted my report, the mainnet configuration for this feature was not set. In other words, this feature wasn't even turned on!
In addition to all of the above, this report was reviewed against the clearly defined terms of our Immunefi program. Based on those terms, issues such as those raised in this report that DO NOT impact block production or consensus are categorized outside of the Blockchain/DLT tier and carry a maximum payout of $50,000.
My response: First, Immunefi has always put the impact of direct fund theft at the very top of its priority list. This is a fact that everyone knows. Second, you changed your bug bounty page after I submitted my report. Here’s the snapshot from November 8, 2025: web.archive.org/web/20251108… . And now, there’s an extra line added to your bug bounty page: “IMPORTANT: Within the Assets in Scope table, the injective-core folder is listed for both Blockchain/DLT and Web/App due to overlap between the two within the same folder. However, for a report to be categorized as Blockchain/DLT, the resulting impact has to be directly involved with the block production process or with consensus failures. All reports not dealing directly with either of these are to be categorized as Web/App.” I’d really like to know when this line was added. and do you really value chain consensus more than users' funds?
We remain committed to fair, transparent, and consistent handling of all reports, and to maintaining the highest standards of security for the ecosystem. Injective has done so since its mainnet inception in 2021 and will continue to do so in perpetuity, always putting builders and security first.
My response: You never even replied to my messages, and now you’re blaming me for not requesting mediation? I can post the original report if you agree. I left many messages, but you haven't replied to a single one.
----------
Finally: Stop making excuses from every angle and trying to use technical jargon to confuse people who aren't developers. That doesn’t work anymore these days. Anyone can just ask an AI to fact-check what both of us are saying. I have no ill intentions toward your project. All I'm asking is for you to be honest and handle this transparently.
Security is paramount at @injective and we take our bug bounty program very seriously.
First and foremost, the figures referenced in the post are entirely misleading. There was no impact realized from this issue. Zero user funds were affected and zero addresses were compromised.
For the stated vulnerability to work in practice, it would require execution of several suspicious transactions that would have an extraordinarily limited impact. Injective has dynamic rate limiting functionalities which are applied automatically based on our live monitoring systems. This functionality has been live on mainnet since last year and is publicly available in our code base.
In addition to all of the above, this report was reviewed against the clearly defined terms of our Immunefi program. Based on those terms, issues such as those raised in this report that DO NOT impact block production or consensus are categorized outside of the Blockchain/DLT tier and carry a maximum payout of $50,000.
If the poster had requested a mediation we would explain to him the dynamic rate limiters and monitoring systems we have in place and why his stated figures are misleading. However, he did not do so. We always follow the procedures set forth by the Immunefi program and expect the submitter to do so as well.
We remain committed to fair, transparent, and consistent handling of all reports, and to maintaining the highest standards of security for the ecosystem. Injective has done so since its mainnet inception in 2021 and will continue to do so in perpetuity, always putting builders and security first.
copper screw retweeted
Replying to @ScrewCopper
Even if you had it too permissive by a little or too strict. It will be a hell of a signal filter.
Highly doubt AI slop would be generated that passes the test.
And if people break the proof system it is a valid submission too (just probably way lower bounty).
At this point there are usable EVM ZK provers so should be feasible.
copper screw retweeted
Hot take: the main reason why contests are dead is because they became free.
- There is less incentive to convince a customer to host a contest when you are not going to make money out of it, especially when there are more lucrative alternatives you can sell.
- There is less incentive to allocate resources to improve a product you don't earn money from.
- In addition to contests becoming free, the % going to judges has been decreasing over time, and meanwhile LLM spam has only gone up which only increases judging cost: this contest only had a disproportionate 4K judging pot, while the contest pot was 500K - and this is a complex codebase with 164K lines of code;
If perhaps 30K was allocated to 2 or 3 senior judges + a cheaper presorter (remember C4 lookouts?) which could spend more time on judging maybe the story would be different.
OMG I just tested this on the beacon deposit contract and found a CRITICAL BUG
🤑brb stealing $161 billion🤑
Codex Security—our application security agent—is now in research preview.
openai.com/index/codex-secur…
copper screw retweeted
Every retweet of this post will equal a dollar given to children with Leukemia.






























