Gustav Hartz
31 posts
Gustav Hartz
@GustavHartz
Founder of https://t.co/TyGg7ZArdg | Smart contract audits
Zurich, Switzerland Katılım Ekim 2020
136 Takip Edilen12 Takipçiler
This is just the tip of where we are heading with LLM’s. 80-90% of critical software probably have undiscovered high severity vulnerabilities
GitHub@github
We are investigating unauthorized access to GitHub’s internal repositories. While we currently have no evidence of impact to customer information stored outside of GitHub’s internal repositories (such as our customers’ enterprises, organizations, and repositories), we are closely monitoring our infrastructure for follow-on activity.
English
Gustav Hartz retweetledi
We are investigating unauthorized access to GitHub’s internal repositories. While we currently have no evidence of impact to customer information stored outside of GitHub’s internal repositories (such as our customers’ enterprises, organizations, and repositories), we are closely monitoring our infrastructure for follow-on activity.
English
So many agentic coding tips on X but still have yet to see any good approaches for using agents for building other long time horizon or deep analysis agents. Issue is that Claude will always suggest adding more stuff to prompt when deletion and generalisation is better
English
Great talk to understand some of the challenges with LLM’s and the fundamental issue with next token prediction. Same root cause that makes AI audit agents perform better with higher token usage and multiple different start conditions
youtube.com/watch?v=ngBraL…
YouTube
English
The audit agent for Supabase is going to be the critical. So easy for agent to shot them selves in the foot building on Supabase. 2/3 of projects probably somehow leaks all customer data by mistake if not more
English
291 submissions in less than 24 hours for @hyperbridge on @HackenProof - The game is changing for bug hunters.
We see two major players:
1. Simple agents scrape, audit, and capture easy findings within 10-30 mins
2. Full-fledged audit systems submitting after 2-12 hours
English
A glimpse into what it takes to build the best AI auditor in the market. Skills are a great first step, but human level audit quality requires thousands of agents and building an agentic harness that gets the best our of them. This is not just asking Claude "plz review"
English
CODE4ARENA is shutting down. Big loss for many bounty hunters and for the wider community. Also, a loss of one of the largest sources of data for AI benchmarking
English
Gustav Hartz retweetledi
Most SaaS pricing is messed up - should drop 5-10x at least. Recently I used Papermark, DocuSign, QR services. Should seriously not cost 40usd pr. month for unlimited QR codes.... Someone build the suite of standard business tools agent first
English
DeepSeek v4 flash pricing in China is unbeatable. This model can power any smart contract auditing solution to 95% accuracy no problem
English
@iosiro_security please reach out. Reviewed one of your recent projects and think we should discuss
English
Just ran some benchmarks on Certiks incident-bench-2026q1. DeepSeek v4 flash is performing at levels of GPT 5.4 in terms of vulnerability detection at 1/10 the price. Cost for Audits just keep plummeting
English
A client of ours got a private preview of Mythos. It found nothing compared to our system. It’s better than nothing, but you can already achieve the same with the models of today and domain knowledge
Claude@claudeai
Claude Security is now in public beta for Claude Enterprise customers. Claude scans your codebase for vulnerabilities, validates each finding to cut false positives, and suggests patches you can review and approve.
English
Spoke to a big bug bounty platform today. The number of reports they are receiving are going up 20% pr. Week! With average signal plummeting they also have to use ai to review. Everyone is trying to earn “easy” money with Claude code right now
English
Gustav Hartz retweetledi
Today, we’re launching the @link wallet for agents. It lets you securely empower agents to spend on your behalf. Your payment credentials are never exposed and you approve every purchase.
link.com/agents
English
Good read. Same principles as for a smart contract audit system
Cloudflare@Cloudflare
Learn about how we built a CI-native AI code reviewer using OpenCode that helps our engineers ship better, safer code. cfl.re/42iCPXe
English
Spoiler: it's not going to hack your project tomorrow
Smart contract scope: Nothing
Blockchain: A couple of minor code defects
Think audit skills are better than this for Web3
English
@rsquare @SafeLabs_ Check out @CecuroAudit - best in the market. Send a dm and we can talk about sponsorship
English
I'm looking for solutions using LLMs/AI/Agents that are enabling a super charged defensive model for teams, products & companies. In particular for @SafeLabs_
Who should I talk to?
English