upvote
I saw a post claiming the issues we're seeing is a result of hasty fixes to AI-discovered vulnerabilities. This seems like the most plausible reason for all of this to me.

Doubly so when we consider companies with external security audits. Those external security firms sometimes have access to better models and larger budgets than the developers responsible for fixing the vulnerabilities they uncover.

reply
Are they doing manual fixes? I have trouble imagining someone who knows how to code and leverages AI properly is breaking something in the process of fixing it. Unless my two assumptions on the people working on this thing are wrong.
reply
Since agents got really at this, I've used them to uncover quite a few verifiable vulnerabilities. Some were easily patchable oversights, but others uncovered flawed architectural decisions which required significant rewrites to fix (done with agents, of course), with a non-zero probability of regressions. I guess your assumption here depends on your definitions of "knowing how to code" and "leveraging AI properly" but it's definitely possible for someone with years of professional experience as a software developer, using the best AI tools on the market, to unintentionally break something when they fix something else.
reply
So all these companies and organizations are just pushing their hotfixes straight to production? Has AI made us forget the very basics of QA? Or are these systems so complex that it's impossible to QA them until they go live for millions of users?
reply
These are huge systems with multiple application, data, and network layers. You can QA the app to death and make sure it works locally and on staging, but then you have firewalls, load balancers, access policies, replication, VPCs, and whatnot. And CloudFormation templates (or whatever IaC you have) is of little help because they are out of sync because someone has definitely made some manual changes in the load balancer config a year go while fighting another crisis.
reply
Deploys have always carried risks, and the speed and checking has always had to be balanced against the benefits of/risks of not deploying now.

You may also have fixed the issue but had a bug in some part of your deployment.

And the more core the bug is (issue in an os) I would wager the larger the set of services you’re likely to need to update is.

reply
The very basics of QA have been forgotten (or more like never learnt in the first place) long before AI.
reply
What do you mean by manual fixes? Have you not seen things break on an upgrade, AI or not?

There is always an extra edge case you could have verified but didn’t, some interaction with reality unaccounted for in testing, complex systems are complex

reply
I suspect a lot of the systems that are getting hastily upgrade haven't been updated or changed in a long, long time. As we know, a lot of systems in the 'real' world outside of tech rarely if ever do updates.
reply
There is a difference between "it's down" and "it's down for maintenance". All we've been hearing is "it's down."
reply
There is currently a large queue with incident response vendors, even if you have a retainer. Some are refusing to take on new customers.
reply
I'm from France and we are also seeing an increase of actual security breaches. So to me the explanation lies in more attaching capabilities than some big US news conspiracy.
reply
[dead]
reply
Let's not forget the mass media's incentive to pounce on the first legitimate story (the unprecedented Microsoft outage), and spin a "foreign hacker" narrative to scare the public and generate clicks.
reply
[flagged]
reply