@category_xyz building @monad · Autodidact · Security · SI/AI/LLM · HTB Top 25 · OSCP · Codebreaker · I post what I'm learning · Sleep is a side channel · 🇺🇸
Joined May 2015
- Tweets1.7K
- Following1.1K
- Followers2.3K
- Likes3.4K
Pinned Tweet
It only took me $12 USD and ~20 minutes to poison an LLM.
Read how I did it here.
ron.stoner.com/How_I_Won_a_C…
🥳 Congratulations! You've successfully time traveled again and all your atoms came through correctly.
The biggest security takeaway from the recent publicized hacks attributed to AI is that you no longer have days or hours to remediate and respond to a compromise.
You have minutes and seconds against a streaming AI session that will pivot, adapt, and continue on when it's time for you to go to bed.
The cat and mouse game continues and currently one of your best tools should be an AI as an additional defender.
You wouldn't eBay an AI datacenter, would you?
ebay.com/itm/880041409723
One of the best SI / AI skills you should be using is a "used car hunter". 🚗
Why are you manually searching sites and looking up VINs and accident reports? Query the whole country at once daily and find your perfect ride.
Quorum certificates?
Where we're going, we don't need quorum certificates.
Post-quantum consensus from @category_xyz
Today we are releasing Quantum Cadence, a post-quantum version of Cadence, our consensus protocol with multiple concurrent proposers. Quantum Cadence uses no quorum certificates, and its safety and liveness do not depend on signatures.
Most modern consensus protocols, Cadence included, rely heavily on signatures and quorum certificates, which a large-scale quantum computer could forge. Simply replacing today's signatures with post-quantum ones would make certificates too large for every validator to send around in a high-performance blockchain. Quantum Cadence instead draws on Forget-IT, a recent signature-free consensus protocol co-developed by researchers at Category Labs.
Quantum Cadence still keeps the overall architecture and core guarantees of Cadence, so our Cadence prototype was easy to carry over: an AI was able to adapt it to Quantum Cadence in a few hours, and an early prototype now runs on an internal test network.
We designed Quantum Cadence to be practical for a high-performance blockchain. In our simulations, its latency is similar to Cadence's, and it scales better with the number of proposers.
Interactive explainer: category.xyz/quantum-cadence.
Full paper: arxiv.org/abs/2610.32803.
Joint work by @MohammadMJalal1, @TobiasKlenze, and @komasito55
Ron Stoner retweeted
gated models are coming regardless - plan to workaround that now
nitter.cf/_jhunsaker/status/2113…
I am okay with Kimi 3.1 deliberately leaving out top-end offensive cyber capabilities from their training data.
Discriminating against use cases is MUCH less bad than discriminating against users. And the world does need all the time it can get to prepare, harden, check old code, (properly) formally verify new code, etc before the full weight of very powerful cyber attacks is upon us.
And if this first big wave of AI risk is not handled well, then we may well get a dystopia of highly locked down models, KYC gating on everything, highly intrusive surveillance on everything, a society where you may be a second or third or fourth-class citizen in terms of model access based on your credentials and nationality.
If you
- got a good night of sleep last night
- have < 10 Slack notifications
- were not besties with PagerDuty or OpsGenie
- woke up and business as usual
then thank your local security professional today
Hey @grok this week and today had a ton of security disclosures and incidents. Let's get an over/under and some lines going on the next protocol or product to get hit. Might as well get some action going and make this interesting.
