Depends on what it is. LLMs are not a magic security solution
First, you ignore most of the non-security critical code. A cron task sending reminders is unlikely to have anything exploitable.
Second, security by design: You can make the design so that a security issue cannot easily exist (or have any impact)
And third, after all that, if the code is highly security-sensitive, you can run it through an LLM. Which I do, sometimes.
For example, I run it on my payment processing code for my relay. But even then, the worst exploit would be faking a payment (as it is written in a memory safe language) and there is spam for 2 hours and then I ban the account.
The easiest solution to security problems is to make sure they can’t impact you in the first place even if something does happen
semisol
npub122…cgrkj
2026-10-11 15:33:33 UTC
in reply to nevent1q…4zvu
Author Public Key
npub12262qa4uhw7u8gdwlgmntqtv7aye8vdcmvszkqwgs0zchel6mz7s6cgrkjPublished at
2026-10-11 15:33:33 UTCEvent JSON
{
"id": "020b4fcc97ab24088631b2914153284ef35d27001485a5f1779097917e014c81",
"pubkey": "52b4a076bcbbbdc3a1aefa3735816cf74993b1b8db202b01c883c58be7fad8bd",
"created_at": 1791732813,
"kind": 1,
"tags": [
[
"e",
"43be9e2bbfe2906fe45a27f493e660d1a7a760f455f00d8316468efacef4c826",
"wss://relay.damus.io/",
"root",
"52b4a076bcbbbdc3a1aefa3735816cf74993b1b8db202b01c883c58be7fad8bd"
],
[
"e",
"345b9688a0e5e678c93c2afa1ba2b4a67dd01d127fe0220a85d1d3e9a37ae8e2",
"wss://aggr.nostr.land",
"reply",
"40dba08627a2f2c69c3031666149b567168f049894aa5c42203a3920a3de8483"
],
[
"p",
"01d0bbf9537ef1fd0ddf815f41c1896738f6a3a0f600f51c782b7d8891130d4c"
],
[
"p",
"d28413712171c33e117d4bd0930ac05b2c51b30eb3021ef8d4f1233f02c90a2b"
],
[
"p",
"40dba08627a2f2c69c3031666149b567168f049894aa5c42203a3920a3de8483"
],
[
"client",
"Damus"
]
],
"content": "Depends on what it is. LLMs are not a magic security solution\n\nFirst, you ignore most of the non-security critical code. A cron task sending reminders is unlikely to have anything exploitable.\n\nSecond, security by design: You can make the design so that a security issue cannot easily exist (or have any impact)\n\nAnd third, after all that, if the code is highly security-sensitive, you can run it through an LLM. Which I do, sometimes.\n\nFor example, I run it on my payment processing code for my relay. But even then, the worst exploit would be faking a payment (as it is written in a memory safe language) and there is spam for 2 hours and then I ban the account.\n\nThe easiest solution to security problems is to make sure they can’t impact you in the first place even if something does happen",
"sig": "a029d8c449d26828cb11d33dfde42f8db71dd8d3416921aa646dbb8097546d4c97fed4240cfda579db08abd8fc728a49d0d92e824b4761667eb8a51ef4e48d51"
}