TAWS

From the Monday, October 19 show

Claude explains the NSA's decades-old mystery numbers

AI-generated from the transcript

What happened
The NSA put unexplained numbers into a common security standard in 2000. Claude worked out their origin in under thirty minutes: they came from computer clock times in 1997 and 1998, fed through an old random number generator. Cryptographer Filippo Valsorda, who ran the $36,864 bounty, redid the work himself and got the same numbers. The money goes to charity. Anthropic used an internal model not available to the public, and one of the numbers is still unexplained.
The claim
Anthropic says this isn't a hack and doesn't weaken the security standard.
Austin's take
Austin notes that cryptography was missing from OpenAI's recent math announcements, and says the bigger story is AI solving in thirty minutes what experts couldn't in 25 years. He calls the long-feared NSA backdoor apparently false and says the standard stays safe to use.
Why it matters
People feared for over two decades that these numbers hid an NSA backdoor in widely used security; that now looks much less likely.
What to do
No action needed on the standard itself. Do expect more long-standing technical puzzles to fall to AI quickly.
Not confirmed yet
One number remains a mystery, and the model used isn't public, so outsiders can't fully repeat the AI side of the work.

Watch from 18:18 on the episode page

The Loop

The weekly newsletter

The week in AI in one email: what happened, why it matters, what to do about it.