Oct 21, 2026, 12:24 PMone finding that i don't have the patience to chase down is that
a) i am 95% sure that the network egress threat classifier has an LLM in the loop that evaluates the requested URL and other contextual metadata as part of the decision, and
b) i am 25% sure that you can cause the classifier to itself receive a classifier refusal by giving it urls that look like prompt injections or biosafety triggers, and rather than either refusing the connection or opening an approval card, it causes a TLS failure.
i have a ~60k/1 week sample confirming (a), (b) is flaky and trying to measure it locks up the instance
![[meta free cash flow reconciliation table over time, I'm not really sure the best way to alt text tabular data, but in summary the free cash flow goes pretty steadily downwards from 15 billion gained in Q3 2024 to being 2 billion LOST in Q3 2026. operating activity revenue increased, but purchases of property and equipment skyrockets until it overtakes revenuem data presented By row, where each number is a quarter. ]
Net cash provided by
operating activities $ 24,724 $ 27,988 $ 24,026 $ 25,561 $ 29,999 $ 36,214 $ 32,226 $ 31,862 $ 32,786
Less: Purchases of
property and equipment 8,258 14,425 12,941 16,538 18,829 21,383 18,997 30,116 33,942
Less: Principal payments
on finance leases 944 411 751 474 545 754 843 962 964
Free Cash Flow $ 15,522 $ 13,152 $ 10,334 $ 8,549 $ 10,625 $ 14,077 $ 12,386 $ 784 $ (2120)](https://media.neuromatch.social/media_attachments/files/117/481/651/386/282/665/original/9d7a447bb682b9ec.jpg)
