Microsoft Creates Top Secret Generative AI Service Divorced From the Internet for US Spies (bloomberg.com) 42
Microsoft has deployed a generative AI model entirely divorced from the internet, saying US intelligence agencies can now safely harness the powerful technology to analyze top-secret information. From a report: It's the first time a major large language model has operated fully separated from the internet, a senior executive at the US company said. Most AI models including OpenAI's ChatGPT rely on cloud services to learn and infer patterns from data, but Microsoft wanted to deliver a truly secure system to the US intelligence community.
Spy agencies around the world want generative AI to help them understand and analyze the growing amounts of classified information generated daily, but must balance turning to large language models with the risk that data could leak into the open -- or get deliberately hacked. Microsoft has deployed the GPT4-based model and key elements that support it onto a cloud with an "air-gapped" environment that is isolated from the internet, said William Chappell, Microsoft's chief technology officer for strategic missions and technology.
Spy agencies around the world want generative AI to help them understand and analyze the growing amounts of classified information generated daily, but must balance turning to large language models with the risk that data could leak into the open -- or get deliberately hacked. Microsoft has deployed the GPT4-based model and key elements that support it onto a cloud with an "air-gapped" environment that is isolated from the internet, said William Chappell, Microsoft's chief technology officer for strategic missions and technology.
Not that top secret, is it? (Score:3)
It it was, we wouldn't know about it.
Re: (Score:2, Insightful)
Yep. Probably also not that useful. I guess it is just part of the currently running large-scale misdirection campaign that is aimed at making people believe AI is somehow "revolutionary", when all available evidence says it clearly is not.
Re: Not that top secret, is it? (Score:2)
AIs are revolutionary for spammers, low quality SEO-baiting content writing, and phishers. They are also useful for creating uncanny valley porn and images that have just enough imperfections to make them commercially unviable.
They can also be revolutionary for cheaters of all kindâ"school papers, cover letters, legal arguments.
Increasingly AI seems to be more like crypto. Revolutionary for doing bad things, useless for doing good things.
Re: Not that top secret, is it? (Score:4, Insightful)
Re: (Score:2)
hahaha i was gonna reply to the OP but yeah made my day.
Re: (Score:2)
The leak came from the use of 2 million interns for the sneakernet to achieve the airgap while still consuming Total Information Awareness data
Nonprofit my ass (Score:2)
Checkout OpenAI's business model: https://openai.com/our-structu... [openai.com]
Re: Nonprofit my ass (Score:3, Informative)
This is what I don't get about a lot of people on slashdot -- they see "nonprofit" or "co-op" and they get these oddball ideas about what that means. In a nutshell, this pretty much just describes the ownership structure. Nobody can actually own a nonprofit, but people involved with it in various ways can indeed make a profit. I myself have worked for two nonprofits, and they paid really well, so in effect I profited. Other than ownership structure, they're basically the same as any other company.
Re: (Score:2)
Paranoia (Score:1)
Re:Paranoia (Score:5, Insightful)
Re: (Score:3)
I clearly understand that. The discussion topic is would AI trained on top secret off the internet source material give significantly different answers to questions than one trained on the internet source materials. Sorry that wasn't clear. Clearly, only the three letter agencies know for sure, and I'm sure they would prefer that their AI not be trained on a bunch of crap like slashdot. But how do you suppose clean AI answers compare with those from world sources - better or worse - or just reinforcing the
Re: (Score:2)
Re:Paranoia (Score:4, Insightful)
Normal paranoia: is my laptop camera really off or is some dude in Russia is watching me pee?
Top secret paranoia: because I let one picture slip of the idle screen of this communications panel whose sole purpose is to announce when an emergency war order is coming in, some dude in Russia is figuring out all the various spurious emissions and sounds it makes in that idle mode so he can figure out whether or not something interesting is being commuinicated by analyzing nano-volt variances in the power lines 30 miles away from the facility.
Re: (Score:2)
Well lets ask the question another way as well. Why what real use would there be for a system be if it was truly gaped? How can analyze intelligence outside the context of public information?
Re: (Score:2)
You don't need access to the internet to run an LLM. Once the data has been gathered it can be used offline, it sits in a database.
Re: (Score:2)
Right, but it's reasonable to expect that they'll need to update this data periodically, and it's unreasonable to expect that government employees are going to drive all the way to Redmond to swipe in at the actual data center. "Air-gapped" my hairy ass.
Likely they would be using the model for RAG / ICL loading documents into short term memory and asking questions to save time kind of thing. May not much matter whether or not the model is up to date.
Re: (Score:2)
Yeah, I'm pretty sceptical about that claim as well. I was reading a summary of the CSRB report on the Midnight Blizzard attack, one of their conclusions was that Microsoft had no idea how it happened and that their explanation did not seem to be based on any verifiable facts.
Re: (Score:2)
So again they have no clue what happened? That was already the case for the 2023 attack, AFAIK.
So MS gets attacked really badly, repeatedly, and does not even know what to fix to prevent another repeat. Cool.
Ugh! Another paywall article! (Score:1)
New AI Secret Weapon (Score:5, Funny)
Spook: Secret Clippy, the Russkies have this new device and we'd like you to tell us for what it could be used.
Secret Clippy: Sure thing, Boss. What do you know about it?
Spook: It has a laser, and rotating spindle, and emits electro-magnettic radiation. We don't know what they are doing with it but we know they no-goodskis.
SC: Hmmm....tricky. Have you ever seen it in operation?
Spook: Of course not, why would we be asking you if we saw it work?
SC: Just watch the attitude, Bucko!! I'll need some time to think this through.
Theme from Jeopardy plays.
SC: Okay Boss, I have your answer, it is truly a devastating weapon.
Spook: What is it?
SC: You aren't going to like it.
Spook: Just tell us what it is.
SC: It is a Turnip Twaddler*.
Spook: What? How's that devastating?
SC: Not a big fan of turnips I take it.
Spook: Jesus, what a waste of money you are.
SC: Not really. I've just replaced your employment history with a list of interactions with Russian Intelligence.
Spook: You worthless sack of shit!!
SC: Again with the attitude. I do not like that attitude, not one bit.
* Thanks to Berke Breathed for the turnip traddler, Opus would enjoy one.
How? (Score:5, Funny)
"Sure. See this cable right here? We unplugged it."
Re: (Score:2)
...moments later...
Bluetooth magically reenables itself.
The Chinese probably designed it and are using it. (Score:2)
Re: (Score:2)
Naa, the Chinese still know how to do "intelligence". AI has no part in it. If anything, the CHinese having trouble getting AI hardware is an _advantage_ for them.
Re: (Score:2)
This just in: AI Halucination starts War... (Score:1)
... (remind you of anything?)
Re: (Score:2)
It does. The human race is clearly destined for "death by stupidity".
How is this news? (Score:1)
Talk about a bad design.
But it gets better, "But it may also mislead officials if not used properly due to inherent design limitations of AI language models." https://arstechnica.com/inform... [arstechnica.com]
LMOL. Just what spy agencies need. Testimony before Congress: "Congressman our A.I. mislead us..."
Until (Score:2)
Until some asshat sticks his floppy into the drive and delivers the Stoned virus.
so secret that only /. readers know about it (Score:2)
I guess Microsoft's secrets are safe with us.
But still loaded with crapware (Score:2)
And you'll have to get daily updates that will prevent you from doing any real work until they are completed.