Post

But who’s thinking about the real victims of that fact-based thought process??!!?! Should the investors’ stakes and huge marketing spends mean nothing to us??! (/s)

Amias Channer

@amias.net

The big problem with the token based a.i. paradigm is that customers have to pay more tokens for security fixes in each request. This incentivizes rapid releasing of low quality because the customer pays your technical debt.

Yash Jain

@earanyash.bsky.social

The interesting bit is neither trick touches an AI model at all. ClickFix works because it's a convincing prompt telling someone to run a command they don't understand, and a helpdesk reset works because verification is still name and DOB. No model needed on either end.

What's happening with GenAI research and Frontier AI(tm) is actively harming cyber defence globally. Lots of orgs and governments have drunk the kool aid and now orgs are unable to defend themselves... from hype from their own vendors.

agrajag9.bsky.social

@agrajag9.bsky.social

Vulnerability assessment and management has become unmanageable.

I lead a team of emerging threat analysts in 3 countries dealing with petabytes of telemetry and real world incidents.. and zero is frontier AI. A slightly better phishing email? Sure. That doesn't drive the incident. Orgs are running around in diapers now with their hair on fire because of hype.

The misaligned incentive is to get you buy new things - tools, processes, consulting - to deal with frontier AI(tm). And for CISOs to secure more budget. That does not lead to better outcomes, when orgs couldn't defend against Advanced Persistent Teenage Ransomware Groups in the first place.