On Thursday I testified before the House Committee on Oversight and Government Reform at a hearing titled “The Federal Government in the Age of Artificial Intelligence.”
The other speakers mostly talked about how cool AI was—and sometimes about how cool their own company was—but I was asked by the Democrats to specifically talk about DOGE and the risks of exfiltrating our data from government agencies and feeding it into AIs.
Anonymous Coward • June 6, 2025 4:22 PM
Will an HKS Fellow pointing out DOGE activities prompt more government attacks against Harvard University?
Ron Helwig • June 7, 2025 7:02 AM
I wouldn’t be worrying about data that government has gathered on me being used by AI or private businesses. The main worry over people’s data is that government has it in the first place. Government has far more ability to cause damage to individuals based on that data.
Guy from Germany • June 7, 2025 11:03 AM
Now, that we just learned about OpenAI storing all prompts and all results, in order to comply with a court order, I think the risk of information leaks should be apparent to everyone.
In principle, I do agree that many governments – including the US – could benefit from a DOGE. However, that would mean doing very cumbersome and unspectacular work rather than throwing some “AI” kids at a pile of government data. Done right, it would take a lot of passion and effort from all people involved.
Anon Mousse Extra Hold • June 7, 2025 11:58 AM
It’s cool guys. AI will just hallucinate Buttle instead of Tuttle.
Clive Robinson • June 7, 2025 12:13 PM
@ Bruce,
When you say,
“I was asked by the Democrats to specifically talk about DOGE and the risks of exfiltrating our data from government agencies and feeding it into AIs.”
Begs the question,
“Is there a smell of burning bridges lingering on the hill?”
Mind you from what has been reported in the MSM just recently Manbaby Hellon Rusk and the Mufina Don the Marmalade Doughnut are not seeing eye to eye any longer…
Also I understand that DOGiE has already cost more than it claimed incorrectly it was saving…
Then there is that “Friends of Russia” incident that has still not been answered…
Of course the question that some may ask is did Hellon get his money back and how much Marmalade is going to get spiteful.
There are stories circulating that NASA did not get a new leader, because of the split, others blame “some immigrant nobodies ever heard of” because he did not get his ring kissed…
All that aside, people do need to stand up and say the expectations stirred up as part of the hype of current AI LLM and ML systems is not going to be delivered by them and that frankly they are very very expensive auto-complete systems that based on probability actually “lower the tone” even when not outputting falsehoods.
And that’s before we talk about how they can oh so simply be used for fraud and other unlawful acts.
As for Corporations getting their hands on the data… I’ve been warning about the likes of Palantir here for quite some time now. They are amalgamating data not just from the US but other nations including the UK. Their aim is to replace human “investigators and analysts” with their computer systems, apparently so far this has not been a success. But as part of the process they got Police and similar to type in a lot of raw data into Palantir’s systems. This data after a little massage and repackaging is being sold to anyone “with a fist full of dollars” without any real checking…
But the AI business plan of Meta and Microsoft and I assume Alphabet as well is as a surveillance tool to gather as much “Personal and Private Information” from people as possible.
I’ve called it the “BE Plan” for,
“Bedazzle, Beguile, Bewitch, Befriend, and BETRAY”’
Because this just shows the steps which are already in full progress and as a result,
“All your privacy will be striped bare”
And endlessly be sold over and over, passing from hand to hand each making their percentage on you.
That’s “the Future in the US” and every where else if the big Silicon Valley Corps can lobby and get the legislation they want…
Steve • June 7, 2025 3:55 PM
@Clive: That’s “the Future in the US” and every where else if the big Silicon Valley Corps can lobby and get the legislation they want…
Given the gross incompetence (almost typed “incontinence” there and now that I think of it, that’d probably fit as well) of the Silly Valley Bros as seen by the DOG (sic) fiasco, I’m not sure that we have all that many worries.
Most of those clowns got in early and got lucky once and are now convinced they’re geniuses when they’re really walking, talking examplars of the Dunning-Kruger Effect with the ability to custom order hats and tee shirts.
Mad DOGE and South Africans, out in the noonday sun. . .
Mr. Peed Off • June 7, 2025 3:57 PM
To spend years constructing an artificial intelligence platform long before the market understood its value or potential applications was repeatedly derided.
Many were skeptical, if not outright hostile. Early investors were skittish; their commitment was often fleeting, at best.
Our results are not and will never be the ultimate measure of the value, broadly defined, of our business. We have grander and more idiosyncratic aims.
Yet we are subtle enough students of history to understand that strength, particularly the type that makes possible the nonconsensual compliance of a foreign adversary, is often, and indeed unfortunately, an essential component of forward progress.
Similarly, our U.S. government revenue increased 45% year-over-year to $373 million in the first three months of 2025, as our software systems for planning and executing special forces and other military operations, and for assessing and selecting targets, has been embraced by the American defense sector.
Clive Robinson • June 7, 2025 11:36 PM
@ Bruce,
You made it into the UK’s “The Register” over your performance…
https://www.theregister.com/2025/06/06/schneier_doge_risks/
Apparently you “played the part of the skunk at the garden party”…
An expression I’d not heard before.
Any way they actually regarded what youl’d said in a mostly positive way.
However they did note one thing,
“Schneier’s message wasn’t what the lawmakers running the hearing wanted to hear – in fact, his testimony goes unmentioned amid the Oversight committee’s AI-boosting summary of the event.”
Shame on them for creating faux-history.
Rontea • June 8, 2025 10:34 AM
Thank you for your thoughtful testimony, Professor Schneier. Far from being the skunk at the garden party, I believe you were indeed the essence and spirit of the gathering itself.
“Hallucinations are real”
Philosophically speaking, the notion of reality is often tethered to perception. Hallucinations, though subjective and divergent from a shared consensus of the external world, possess a certain authenticity within the mind experiencing them. As Descartes famously posited, “Cogito, ergo sum”—”I think, therefore I am”—suggesting that the act of thought confirms existence. In this light, the vividness and impact of hallucinations on an individual’s consciousness grant them a form of reality, not as external fact but as internal truth. They are real in the realm of subjective experience, shaping emotions, beliefs, and even behaviors, thus intertwining with our understanding of what it means to perceive and exist.
Clive Robinson • June 8, 2025 11:24 AM
@ Rontea, Bruce, ALL,
With regards,
“In this light, the vividness and impact of hallucinations on an individual’s consciousness grant them a form of reality, not as external fact but as internal truth.”
You might want to consider,
https://www.ucl.ac.uk/news/2025/jun/brain-mechanisms-distinguish-imagination-reality-discovered
Basically fMRI was used whilst volunteers were shown various very noisy images. They were prompted as to what the faint image should be and concentrate on seeing it and describe what they see.
However half the time the images were just noise and had no faint object image in them.
The fMRI picked up brain activity and from this the researchers determined where in the brain people perceived reality from imagination.
Rontea • June 8, 2025 1:45 PM
@Clive Robinson
Thank you for the link. The notion that simulating reality can sometimes feel less absurd than living it, speaks to the power of imagination and technology in crafting our experiences. It’s fascinating to consider how advances in our understanding of brain functions, like the workings of the fusiform gyrus in distinguishing perceived reality, might shape future advancements in mixed reality.
Leslie • June 12, 2025 3:53 AM
Re “Those erroneous dates were an artifact of issues with date-handling in COBOL, a software language used in the mid-20th century to create many of SSA’s data
systems.”
Please don’t abuse COBOL this way. I know it’s fashionable to denigrate the language, but the issue with dates has nothing to do with the language, and everything to do with the computer resources available in mainframes’ early years.
For instance, before IBM developed the System/370 series with their virtual storage memory system, their System/360 series machines’ memory size ranged from just 4K to 9M, tiny compared to even today’s cell phones. The System/370 series real memory size ranged from only 16M to 64M, still pretty small.
The main memory sizes of systems from other vendors during that era were of comparable magnitude.
Because this limited program sizes, spaces saving methods like trimming the year portion of dates to two digits was almost universally practiced, and that is what led to the date artifacts you mentioned; it had nothing at all to do with COBOL itself. (In fact, in many ways, COBOL, being a language that provides relatively static data structures, is more reliable than modern languages because it doesn’t easily allow security errors like string overflows.)
So, date artifacts from early “legacy” systems have nothing at all to do with COBOL or any other language, but everything to do with storage constraints.
Dip Stick • June 14, 2025 4:11 AM
Secure everything with AI and then install TCPDump absolutely everywhere possible. Run the entire enterprise with a 22 year old who does not know what TCPDump is (just for fun). 😉
You Too • June 19, 2025 3:00 PM
Bruce’s opening remarks and excellent analysis of the DOGE dangers begins here:
https://www.youtube.com/live/wKkk-uWi7HM?si=Qz_liXaxg0sZFZW1&t=4476
Subscribe to comments on this entry
Sidebar photo of Bruce Schneier by Joe MacInnis.
lurker • June 6, 2025 3:04 PM
If I were a nation state adversary I might concentrate my initial kinetic attacks first against the cable distribution points, then against the data centres, worldwide. No cloud, no AI, could be a significant short to medium term disability to business and govt. No data to steal.