I am pushing myself to learn more in public this year and am excited to share my first ever writeup about a vulnerability I found in a verification system used by @worldnetwork.
I'll also share a script for finding similar bugs #bugbountytips
1/n
medium.com/@gonzo-hacks/the-…
The video of the panel I moderated at DEF CON is live!
It’s no secret that AI has changed the cybersecurity landscape, and it’s reshaping the operations and economics of bug bounty.
Hear from frontline program managers on what’s changed, how it’s changed, and what the future of bug bounty looks like.
youtube.com/watch?v=gWNif2Ia…
Working with AI requires perpetually operating at the edge of sanity. You have to recalibrate your mental model of AI capabilities and your own job description every few months, without getting delusional about seemingly infinite productivity gains, or feeling threatened by losing what you thought was the uniquely human part of your job, or obsessing over the perfect setup, or losing your mind over the clumsiness and jaggedness of agents, or ...
Maybe AI literacy needs to be less about prompting and more about improving emotional regulation.
Half of cybersecurity Twitter is asking you to chill. The other half wants you alarmed. Here's a primer:
Team Chill: "X has barely ever/never happened, resource needs are too high & brittle, so chill & focus on basics"
Often people with deep career experience, a lot of it wise & we have much to learn from them. But some of the heuristics they use to grade threats are indexed to a stale cost model.
For example, difficulty & complexity of attacks just don't mean the same thing that they did 5 years ago. Not for attribution (complex no longer = must be a gov!), and not for risk (my thing is obscure & complex & therefore safer than the popular thing).
At the same time, radically new things are happening around AI and cybersecurity and these folks are trying to understand a radically new thing through deep experience.
Team Threat: "X is now possible, so this IS a threat"
Often AI native ppl or practitioners seeing how LLMs are shifting what we thought was possible. And so many of them are drawn to stuff that is shiny and/or stuff that was historically in the low probability, high impact quadrant. But suddenly the impossible is...foreseeable?
Something that felt like an impossible resource spend or hyper esoteric proof of concept is now... scaffolded on $24 dollar zero days with fewer resources & more messy parameters.
So the challenge is: bringing those old, wise intuitions to bear even as some parts of the world are changing underfoot, fast. Right now, so much of the discourse is prospective & anticipatory.
Team chill is probably right about the median attack, and Team Threat is probably catching tail distribution but may be wrong on how soon it is coming.
It's boring to say truth is in the middle, but I think truth will probably emerge the more we are collectively working on emerging attacks & problems with good evidence & transparency.
This is also why why the recent threat reports (e.g. Anthropic's Sept 2026 monster threat report) matter so much. We desperately need empirical grounding to help suss out blind spots.
I claimed the "toormund" handle on @ThePirateFace 🏴☠️
Where Hugging Face AI models never die, and are immortalized as torrents.
Claim yours 👇
pirateface.co/claim?ref=toor…
Omg what! The Hugging Face deal price was the code for the hugging 🤗 emoji. Did someone leave money on the table to make $12.9303 happen?? businessinsider.com/nvidia-h…
5) BigCo security researchers will be loathe to publish jailbreak research or to share it with the government, lest they get blamed for an overreaction that sets back the entire US industry. Likewise, there is a huge upside for PRC jailbreaks of US models now.
Patch the Planet is our effort to help open source maintainers move from security findings to merged fixes.
We’re working with Trail of Bits, HackerOne, Calif, researchers, and maintainers to bring Codex Security and advanced models into the remediation process, with human review at the center.
New on the Engineering Blog: The access and permissions we grant agents should evolve with their capabilities. In our own products, we set these parameters through sandboxing, which limits the scope of any potentially destructive actions.
Read more: anthropic.com/engineering/ho…
In this talk, @DaneSherrets and @Shlibness break down how modern AI systems are being hacked in the real world and what bug bounty hunters and organizations can do to keep up.
youtu.be/LDNvqiUnF5I
Conversations around Mythos center around a potential tidal wave of vulnerabilities. It’s easy to jump to worst-case scenarios.
But that’s not the full picture.
HackerOne’s @DaneSherrets offers a more grounded perspective: the same AI capabilities accelerating discovery are also accelerating defense. Vulnerabilities can be found faster and fixed faster.
The real advantage? It belongs to the teams who act first.