Meta's AI safety system defeated by the space bar
'Ignore previous instructions' thwarts Prompt-Guard model if you just add some good ol' ASCII code 32 Meta's machine-learning model for detecting prompt injection attacks – special prompts to make neural networks behave inappropriately – is itself vulnerable to, you guessed it, prompt injection attacks.…
#theregister #IT
https://go.theregister.com/feed/www.theregister.com/2024/07/29/meta_ai_safety/
The Register /
npub10c…f4lpy
2024-07-29 21:09:02
Author Public Key
npub10czyvexf06s7m58632xn8ddk5pe5arkw8vfxsp7qzw3petnx009saf4lpyPublished at
2024-07-29 21:09:02Event JSON
{
"id": "a1a893aad8eb9611df7a170ec60689aa4535ec58aae5a6a1660d8a3debf7afa3",
"pubkey": "7e044664c97ea1edd0fa8a8d33b5b6a0734e8ece3b126807c013a21cae667bcb",
"created_at": 1722287342,
"kind": 1,
"tags": [
[
"t",
"it"
],
[
"proxy",
"https://geeknews.chat/@theregister/112871823289564227",
"web"
],
[
"t",
"theregister"
],
[
"proxy",
"https://geeknews.chat/users/theregister/statuses/112871823289564227",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://geeknews.chat/users/theregister/statuses/112871823289564227",
"pink.momostr"
],
[
"-"
]
],
"content": "Meta's AI safety system defeated by the space bar\n\n'Ignore previous instructions' thwarts Prompt-Guard model if you just add some good ol' ASCII code 32 Meta's machine-learning model for detecting prompt injection attacks – special prompts to make neural networks behave inappropriately – is itself vulnerable to, you guessed it, prompt injection attacks.…\n#theregister #IT\nhttps://go.theregister.com/feed/www.theregister.com/2024/07/29/meta_ai_safety/",
"sig": "9947468a70d6137740f2e75d0b80d76633a95c1a073819b8adb4485a6835b171d32b9e9b01329d65cdacbe81cd42d0a43cf02aacc800c6958c5b9c77ce62e5c5"
}