My director of security engineering investigates every new vuln for the sole purpose of determining how aggressively he can shitpost about it.
10/10 vulnerability management strategy.
Wowww. EVERYONE says yes. I dunno… I am skeptical. Do IR companies only accept a limited number of clients on retainer then?? If there’s a widespread issue that’s causing them to be so underwater they can’t accept non-retainer clients, how attentive are they going to be to you?
I say keep a list of the 10 least well-known companies on the approved insurance list, their specializations, and their emergency line. Dial as needed. I’ve worked with a “big-name” IR firm. Not impressed.
Except there are idiotic regulatory restrictions from insurance companies that your IR firm is not allowed to perform any other security services for you. You can’t actually put a firm on retainer that knows your systems. But yeah… I guess at least the paperwork is already signed.
One of the great things about the influx of vulns is that it used to be a whole thing to mobilize the troops. Now, it’s just another day at the office. Great way to operationalize an important process.
The economics of the exploit market have changed. There used to be an incentive to sit on a pile of vulns patiently waiting and working years to get full RCE from them. You’d make 6-7 figures instead of 5. Now, you just gotta sell that shit as soon as you find it.
The agent violates an instruction because it interpreted it in some weird way. You add a more abstract instruction to try to guide it. It violates that instruction. You add another. 3 hours later you're writing shit like "do not interpret the word 'is' to mean 'is not'".
@IceSolst Because you don’t get increased output from a new technology or process overnight. It takes time to figure out how to properly take advantage of it. Especially at scale in a large organization. Also… some devs will 5x their output and some will spend 5x less effort.
Open source all exploits as soon as you have them, I think that’ll be the most significant acceleration of remediation
Also remove the cyber guardrails on all models
I didn’t know about the financing and Kress’ apparent disagreements with it. But even then… his investment in open models and the potential for nvidia to dethrone the big players is still a highly plausible existential threat to their companies. So maybe they lose financing in the short term… but they cripple a longer term threat of being killed by the competition. Will be interesting to see how it evolves. Does nvidia stop the financing? Does the hugging face acquisition go through??
>be ai company
>warn everyone that AI will hack the planet
>build increasingly powerful model
>eval how good it is at hacking the planet
>be me, cybersecurity person trying to defend my company’s shit
>ask ai company if they run defensive evals
>"uuuuh… mostly offensive"
>point out that some of us are trying to defend
>also why are you training the monster to be good at the exact thing you’re telling us to be terrified of
>"actually… im sure we do defensive evals too"
>crisis averted
“Each crop of unseemly gossip, thus harvested, becomes the seed of more, and, in direct proportion to its circulation, results in a lowering of social standards and of morality. Even gossip apparently harmless, when widely and persistently circulated, is potent for evil. It both belittles and perverts. It belittles by inverting the relative importance of things, thus dwarfing the thoughts and aspirations of a people. When personal gossip attains the dignity of print, and crowds the space available for matters of real interest to the community, what wonder that the ignorant and thoughtless mistake its relative importance.” - Warren & Brandeis 1891
@techspence posted something on another platform that hit home:
For every unimportant thing that is fixed or made a priority, something truly critical gets moved down the list.
Sometimes people have asked me “what’s the harm when media frenzy, marketing buzz, or knee-jerk reactions cause a topic that is not really a thing to take center stage (think agent swarms in toasters).
The harm is that all of those conversations, thoughts, and actions were about something unimportant, and were not about Winter2026! as a password, unmanaged devices, misconfigurations, users running as local admin, password stored in the clear, and so on…
Every time we make something unimportant take center stage, what should be center stage does not, and the attacker gains an advantage. It is like skipping a turn in a card game. Make a habit of skipping turns, and you will lose.
This has also been pointed out repeatedly by @ZackKorman, @UK_Daniel_Card, @Shammahwoods, @IceSolst, @christian_tail, and so many others here.