hckrnws

Not a fan of future products being announced as if they are here but are basically is still in "Internal Research" stages. I'm not sure who this is really helping? except creating unnecessary anticipation which we kinda all know are in this loop lately of "yes it works great, but".

I'm optimistic that it's easier to find/solve vulnerabilities via auto pen-testing / patching, and other security measures, than it will be to find/exploit vulnerabilities after - ie defense is easier in an auto-security world.

Does anyone disagree?

This is purely my intuition, but I'm interested in how others are thinking about it.

All this with the mega caveat of this assuming very widespread adoption of these defenses, which we know won't be true and auto-hacking may be rampant for a while.

In open source codebases perhaps, either because big tech would be generous enough to run and generate PRs(if they are welcome ) for those issues.

In proprietary/closed source it depends on ability to spend the money these tools would end up costing.

As there is more and more vibe coded apps there will be more security bugs because app owners just don’t know better or don’t care to fix them .

This happened when rise of Wordpress and other cmses and their plugin ecosystem or languages like early PHP or for that matter even C opened up software development to wider communities.

On average we will see more issues not less.

In general, most modern vulnerabilities are initially identified with fuzzing systems under abnormal conditions. Whether these issues may be consistently exploited can be probabilistic in nature, and thus repeatability with a POC dataset is already difficult.

That being said, most modern exploits are already auto-generated though brute-force, as nothing more complex is required.

>Does anyone disagree?

CVE agents already pose a serious threat vector in and of itself.

1. Models can't currently be made inherently trustworthy, and the people claiming otherwise are selling something.

"Sleeper Agents in Large Language Models - Computerphile"

https://www.youtube.com/watch?v=wL22URoMZjo

2. LLMs can negatively impact logical function in human users. However, people feel 20% more productive, and that makes their contributed work dangerous.

3. People are already bad at reconciling their instincts and rational evaluation. Adding additional logical impairments is not wise:

https://www.youtube.com/watch?v=-Pc3IuVNuO0

4. Auto merging vulnerabilities into opensource is already a concern, as it falls into the ambiguous "Malicious sabotage" or "Incompetent noob" classifications. How do we know someone or some models intent? We can't, and thus the code base could turn into an incoherent mess for human readers.

Mitigating risk:

i. Offline agents should only have read-access to advise on identified problem patterns.

ii. Code should never be cut-and-pasted, but rather evaluated for its meaning.

iii. Assume a system is already compromised, and consider how to handle the situation. In this line of reasoning, the policy choices should become clear.

Best of luck, =3

I've also thought this for scam perpetration vs mitigation. An AI listening to grandma's call would surely detect most confidence or pig butchering scams (or suggest how to verify), and be able to cast doubt on the caller's intentions or inform a trusted relative before the scammer can build up rapport. Security and surveillance concerns notwithstanding.

I wonder if we're going to end up in an arms race between AIs masquerading as contributors (and security researchers) trying to introduce vulnerabilities into popular libraries, and AIs trying to detect and fix them.

Why would it be like that instead of the way we already handle low-trust environments?

Projects that get a lot of attention already put up barriers to new contributions, and the ones that get less attention will continue to get less attention.

The review process cannot be left to AI because it will introduce uncertainty nobody wants to be held responsible for.

If anything, the people who have always seen code as a mere means to an end will finally come to a forced decision: either stop fucking around or get out of the way.

An adversarial web is ultimately good for software quality, but less open than it used to be. I'm not even sure if that's a bad thing.

What I'm suggesting is: what if AIs get so good at crafting vulnerable (but apparently innocent) code than human review cannot reliably catch them?

And saying "ones that get less attention will continue to get less attention" is like imagining that only popular email addresses get spammed. Once malice is automated, everyone gets attention.

Significantly easier to detect than create? Not quite NP, but intuitively an AI which can create such an exploit could also detect it.

The economics is more about how much the defender is willing to spend in advance protection vs the expected value of a security failure

4.5 million lines of code for one fix is impressive for an LLM agent, but there's so little detail in this post otherwise. Perhaps this is a tease to what will be released on Thursday...

That's how I read it at first too, but I think the more probable interpretation is that it was a fix to a project that has 4.5M lines of code.

oh, that would definitely make more sense.

Can we just flag this since it’s not actually a thing available to anyone?

So it is a secret tool, they will "gradually reach out to interested maintainers of critical open source projects with CodeMender-generated patches", then they "hope to release CodeMender as a tool that can be used by all software developers".

Why is everything in "AI" shrouded in mystery, hidden behind $200 monthly payments and has glossy announcements. Just release the damn thing and let us test it. You know, like the software we write and that you steal from us.

DeepMind = not available for use

It's lost its charm.

what an annoying page

pointless videos, without enough time to read the code