AI blew open software security, now OpenAI wants to fix it with an agent called Aardvark

After helping expand the modern software attack surface with the rise of AI services prone to data poisoning and prompt injection, OpenAI has thrown a bone to cyber defenders. The maker of ChatGPT on Thursday announced that it is privately testing Aardvark, an agentic security system based on GPT‑5. "Aardvark represents a …

[itsborken]

Brilliant

Train OpenAI on your proprietary software--what could go wrong? I'd check those T&C very closely.

[DS999]

Re: Brilliant

I'd check those T&C very closely

You would trust their T&C's if they pinky swore they absolutely won't collect any information about your proprietary software to help train ChatGPT v6? No point in "checking" something if you can assume the counterparty is inherently untrustworthy.

They have been running roughshod over copyright internet-wide since day one. Unless you're the size of Apple or Microsoft your lawyers will crumble before theirs even if you can prove they violated that pinky swear.

[amanfromMars 1]

EDITorially Creative CodeXSSXXXX Leaders for Tales of Future Success Guaranteed

And what does Aardvark answer whenever asked if it and IT can be tasked to exclusively provide proprietary executive command and COSMIC* Top Secret Mk-Ultra type control of Universal Self-Serving Instruction Sets that also Guarantee Brave NEUKlearer Virtually Real World Order Resets ...... for that is what it is sure to encounter with acceptance of any of those many always open invitations to validate and/or scramble such as would be is Extremely Disruptive IT and Creative CodeXSSXXXX.

It and IT isn't going away you know whenever everything everywhere, both here and over and out there, is rotten ripe ready for the picking and bottling/processing and packaging/AISteering/Mega and MAGA MetaDataBase Mining.

COSMIC* .. Control Of Secret Materiel in an Internetional Command

[Dan 55]

This is just Atlas for programmers

OpenAI wants to get all the web pages and all the code.

[DarkwavePunk]

Aardvark

Bet they chose that name just to be on the first page of the Yellow Pages directory listing. No I'm not old - what gave you that idea?

Also it just seems like slop eating it's own tail. AI Jormungand without the Viking cool factor.

[abend0c4]

Funny old world

I remember when there was serious talk of proving programs correct.

And now we are to put our faith in indeterminacy.

[DarkwavePunk]

Re: Funny old world

Gödel had something to say about that. Have an upvote anyway. Everyone needs to read Gödel Escher Bach. Well, if you like brain damage through formal logic that is...

[Anonymous Coward]

Re: Funny old world

"Everyone needs to read Gödel Escher Bach."

Actually worthwhile period. As much about pondering the possibility and limitations of man made technology ever having intelligence _{[circumlocution to avoid the dreaded AI words]} as about formal systems and logic.

The actual mathematics is fairly digestible by anyone with high school algebra. The Lem extracts are amusing in themselves.

In the context of proving programs correct I am not sure Gödel's incompleteness theorems are a practical problem. Correctness is always with respect to a formal specification which presumably for a real world problem is satisfiable.

The operations on a computer are (mostly) well defined and can be the basis for reasoning — Predicate Transformer Semantics is one approach; Denotational Semantics another.

Unfortunately even for fairly small programs the effort required to create the required proofs (reasoning) even with the assistance of artificial reasoning (theorem proving) software, is enormous.

Understandably formally correct software is largely restricted to critical systems (or critical components of) where the effort and cost can be justified.

[DrSunshine0104]

For Free?

This kind of feels like a racket.

A gangster turning his goons loose on a city, but puts his hand on your shoulder and tells you he can protect your from the rampant crime if you pay him some money and let his accountant take a look at your books.

[Snowy]

Re: For Free?

If it is free you are the product?

[ecofeco]

Re: For Free?

It's extortion and always was.

It's really the only thing corporations offer these days.

[munnoch]

"LLM-powered reasoning and tool-use to understand code behavior"

Bollocks.

<EOM>

[Anonymous Coward]

Re: "LLM-powered reasoning and tool-use to understand code behavior"

And makes toasted muffins too.

As far as I understand AI/LLM does neither reason in the logical sense nor understands anything — any more than egrep — at best AI recognises form in some sort of weighted graph but doesn't associate anything approaching semantics with the recognised forms or can consistently tranform one recognised form into another form while preserving the senantics (if there were any) which be the essential basis for reasoning let alone understanding.

Honestly AI is full of one handed typers; tossers.

[DaveTP]

Should be safe..

As my Parents always said ,Aardvark never hurt anybody.

Hat, Coat thanks !

[Peter Gathercole]

Re: Should be safe..

Don't know whether there were earlier precedents, but I know that that line appeared in the Goon Shows on the BBC Wireless box.

The time I remember it, I think it was said by Neddie Seagoon, but could have been Grytpype-Thynne, but that is a moot point.

Moriarty, stop pointing that moot at me!