NSA Spends Billions Testing AI Models, Far More Than Previously Known
Classified Estimates Show the NSA Is Paying Billions to Test AI Models
The National Security Agency is spending billions of taxpayer dollars this year to evaluate advanced AI models, according to classified estimates shared with lawmakers. The figure is far higher than earlier federal projections, which put AI oversight costs in the tens of millions. The NSA's Artificial Intelligence Security Center tests frontier models for national security vulnerabilities, and the expense—driven by computing power and scarce talent—is fueling debate over whether AI companies should help foot the bill.
If we want the government to be equipped to assess these systems, we need to fund the computing resources and expertise that requires. I think of these tests as public goods, but there are reasonable questions about if taxpayers should foot the bill.
- paimapi
I feel like any AI regulation passed wouldn't have enough teeth to rein in the NSA because not even the Fourth Amendment seems applicable to the agency. let's not forget the ex-post-facto legalization of mass wiretapping and surveillance [0] with both establishment political parties supporting it on 'national security' concerns [1]. whatever monitoring or testing they're doing today will be similarly hand-waved away in the name of security theater
this is just how the empire has always operated - we have always been at war with Eastasia
[0] https://en.wikipedia.org/wiki/NSA_warrantless_surveillance_(...
[1] https://www.eff.org/deeplinks/2012/10/fact-check-obamas-misl...
- rdtsc
I guess I would be more surprised if the headline said "turns out NSA dropped the ball on AI and haven't even fired up a single LLM instance yet".
It's mostly about "testing" and "billions". But I am sure "testing" here is in quotes just like I wrote it. They are not going to get an itemized list unless it's coming from Snowden 2.0
- We "tested AI" on this Russian nuclear silo command and control center. It worked out pretty ok.
- bhouston
"Testing"? Sure, Jan.
They would have to be incompetent if they were not using them both for security penetration as well as social engineering / election meddling as everyone else is.
- jawiggins
An underappreciated aspect of LLMs is how good they are at zero-shot classification. I'd expect this is extremely good at monitoring communications streams, particularly in foreign languages.
- hosel
Why wouldn’t they be? How is this surprising. A drop in the bucket.
- robinsinclair
I seem to recall a certain former NSA affiliated person speaking on the conjecture of NSA Full Capture from fiber optic taps at a HOPE conference many many years ago. The statements alluded to automated custom robotic “drive shufflers” for even the tape and drive libraries already in use at that time, and heavily implied that in some cases exotic storage was being handled by Rube Goldberg style computer assisted robotically serviced storage libraries that would make even HAL jealous.
- vjvjvjvjghv
What do they test? In general, I don't understand the whole "AI safety" stuff. What are the scenarios we need to be protected from and how is this going to be accomplished?
- rand846633
Does anyone here happen to know the ballpark number of how much the three letters agencies are currently paying OpenAI/antropic/who ever per year/month..?