TanyaKaroli on Nostr: Gary Marcus’ most recent substack included this long tweet, detailing how an AI ...
Gary Marcus’ most recent substack included this long tweet, detailing how an AI agent was easily jailbreaked to plan for and hire a hitman on the dark web. It was a red team hacking exercise performed in a safe environment so no harm done — bit releasing AI agents carries enormous risks. This needs to be regulated massively.
#generativeAI #SecurityRisks
(Text of tweet in next toot since it is so long)
Published at
2025-01-16 05:08:46Event JSON
{
"id": "705d9bd0d823473d1b0e293fda97894da8a3d053436e27ae53f000b6e6bc4b4c",
"pubkey": "bdc4be75258983e1ff37b249751c0384d021845cc8af4f0df832a1926933862c",
"created_at": 1737004126,
"kind": 1,
"tags": [
[
"t",
"generativeai"
],
[
"t",
"securityrisks"
],
[
"imeta",
"url https://media.expressional.social/media_attachments/files/113/836/290/308/863/757/original/fc90287cccb08521.png",
"m image/png",
"dim 914x1536",
"blurhash UkPGjXM{-;of~qj[M{t79Fayt7WB%Mj[Rjjt"
],
[
"proxy",
"https://expressional.social/users/tanyakaroli/statuses/113836302422848734",
"activitypub"
]
],
"content": "Gary Marcus’ most recent substack included this long tweet, detailing how an AI agent was easily jailbreaked to plan for and hire a hitman on the dark web. It was a red team hacking exercise performed in a safe environment so no harm done — bit releasing AI agents carries enormous risks. This needs to be regulated massively. \n#generativeAI #SecurityRisks \n(Text of tweet in next toot since it is so long)\n\nhttps://media.expressional.social/media_attachments/files/113/836/290/308/863/757/original/fc90287cccb08521.png",
"sig": "faa2b59992b0d43f1a721b07c35911c37b974cc3f13c79721e0aef26a9fcd867ccf3554d6f2eb452329dfa4a64943b4f4cb90ec8de743efbcc56c457f1729b61"
}
