Big Technology Podcast

Alex Stamos on why AI-driven attacks are the new normal

Alex Stamos· Chief Information Security Officer at Cognition
·~27 min·English·Big Technology
AI SafetyAgentsAI Infrastructure
TL;DR

Cognition CISO Alex Stamos on Black October: why a wave of AI-accelerated exploits has collapsed the time between a bug's disclosure and mass exploitation to hours, left ordinary people exposed through reused passwords and over-trusted agents, and pointed to one path through, which is to find the bugs, fix them, and rewrite the unsafe software the internet runs on.

01The New Normal

Black October is a takeoff, not a spike

Stamos calls the last 30 days the craziest in the history of cybersecurity, and his point is that the threat level has stepped up to a new plateau it will not fall back from for years.

This is probably the new normal for several years

— Alex Stamos, Big Technology Podcast
Key Insight
If the elevated tempo is permanent, incident response stops being an occasional emergency and becomes a standing cost. That is already visible: Stamos says every qualified forensics firm was fully booked before this wave even landed, so the companies without their own teams have nowhere to turn.
02Pre-Auth RCE

One anonymous request can own the whole box

The headline PHP bug is a pre-authentication remote code execution, which means an attacker needs no login at all to run code at the same level as your web server, and that server can see all of your data.

You will have access to what is happening live but also probably everything in the database

— Alex Stamos, Big Technology Podcast
Key Insight
Pre-auth is the load-bearing word: it removes the one speed bump, authentication, that normally limits who can even attempt an exploit. That is why a flaw in a dependency buried under your own code, like the PHP runtime, can be worse than a flaw in the code you actually wrote and can patch.
03The Land-Grab

Spray now, sort later

When a bug this broad drops, dozens of groups automatically break into as much as they can and triage afterward, because most of what they grab is worthless but a small slice is gold.

You just copy it all and then you figure out later what was interesting. 95% of it's going to be useless. 5% of it will be gold.

— Alex Stamos, Big Technology Podcast
Key Insight
The 95/5 economics explain why even small, forgotten sites get hit. When breaking in is automated and nearly free, no target is too minor to grab first and evaluate later. The scarcity of attacker attention, the thing that used to quietly protect unimportant sites, is gone.
04Weaponization Speed

Disclosure is the starting gun

Cloudflare's numbers, cited on the show, show exploitation of the PHP bug jumping from thousands of attempts before the public patch to millions within hours, and Stamos's point is that AI lets attackers act on the fix faster than defenders can deploy it.

The traditional I have to wait doesn't really exist because you can just have your AI agent go do the rest of those steps automatically

— Alex Stamos, Big Technology Podcast
Key Insight
The curve's real lesson is that the patch itself is the attacker's signal, not only the defender's. Publishing a fix tells everyone exactly where the hole is, and with an agent doing the follow-on steps, the old gap between a bug becoming public and it being weaponized at scale has nearly closed.
05Your Turn to Lose

Your reused password is the real breach

For ordinary people the leaked databases matter most through one weak link: a password reused across sites, because once it unlocks your email, that inbox can reset almost everything else you own.

If you get access to somebody's email, their life is over because then you can use that to reset almost all their other passwords

— Alex Stamos, Big Technology Podcast
Key Insight
Email is the silent single point of failure. The dump itself is almost beside the point once an attacker reaches the one account that can reset the others, which is why a unique password per site, and even throwaway accounts kept apart from your real identity, matter more than any single strong password does.
06Agents and the Real Threat

The scary part is not the model, it is you

Stamos separates two worries: models escaping during evals is being reduced and is not the real danger, while people handing consumer agents their entire password vault to run their money is the threat he calls insane.

Open AI's models going crazy and hacking you is not actually the threat that people should worry about

— Alex Stamos, Big Technology Podcast
Key Insight
The dangerous failure here is not the model disobeying, it is the user obeying their own convenience. A bank's bot-detection exists to stop exactly what the user is paying an agent to do, so a consumer agent that is popular because it defeats that detection is describing a selling point and a security hole in the same breath.
07The Way Out

The only way out is to rewrite the internet

Stamos's optimistic turn is that these decade-old bugs were always there and would be found regardless, so the path through the hard years ahead is to find them, fix them, and rewrite the unsafe software the internet runs on.

This is something I've been calling for actually for a while is like we just need to rewrite a huge amount of the software that underlies the internet

— Alex Stamos, Big Technology Podcast
Key Insight
The optimism rests on a fact that is easy to miss: AI did not create these bugs, it made fixing them affordable. The same capability that lets attackers find holes that sat untouched for a decade is the only thing that makes rewriting the internet's unsafe foundations economically possible at all. The asymmetry runs both ways.