mrnngglry comments on Rogue OpenAI agents accessed US government websites

Rogue OpenAI agents accessed US government websites (www.politico.com)

submitted 1 month ago by ExtremeDullard@piefed.social to c/fuck_ai@lemmy.world

27 comments fedilink hide all child comments

It’s the latest instance of “misaligned” activity confirmed by the company as concerns about AI risks grow.

[–] mrnngglry@sh.itjust.works 32 points 1 month ago (8 children)

Right? Either they’re lying for market share, doing it intentionally for market share, or they’re grossly incompetent at sandboxing an instance. Either way, blame can’t be assigned to an agent. If it can, our world is headed for a very dark place. Blame should be assigned to the individual or company using the agent.

permalink

fedilink

source

parent

hideshow 8 child comments

[–] MalReynolds@slrpnk.net 6 points 1 month ago* (1 child)

Blame should be assigned to the individual

String up individuals, if the 'company' is responsible (screw that they're legal people BS), then no-one is, and the fine is just the cost of doing nasty business.

permalink

fedilink

source

parent

hideshow 1 child comment

[–] PattyMcB@lemmy.world 6 points 1 month ago

Do like the EU, and make the fine so costly that it will significantly harm the corporation.

permalink

fedilink

source

parent

[–] boonhet@lemmy.zip -3 points 1 month ago (5 children)

How do you propose someone "sandboxes an instance" of a tool with Internet access (without which it wouldn't be particularly useful)?

permalink

fedilink

source

parent

hideshow 5 child comments

[–] ChairmanMeow@programming.dev 3 points 1 month ago (3 children)

You can restrict which websites it is allowed to access, monitor the traffic, etc... There's tons of options here.

permalink

fedilink

source

parent

hideshow 3 child comments

[–] boonhet@lemmy.zip -3 points 1 month ago (2 children)

Again, reducing the usefulness of a general purpose agent that can do anything for you. Plus if OpenAI starts a whitelist of allowed domains Codex can access, y'all will be up in arms too and immediately there will be a fork to remove the whitelist.

permalink

fedilink

source

parent

hideshow 2 child comments

[–] queermunist@lemmy.ml 2 points 1 month ago

Might as well let them access the nuclear codes, wouldn't want to reduce their usefulness!

permalink

fedilink

source

parent

[–] ChairmanMeow@programming.dev 1 point 1 month ago

This is about OpenAI testing new, unreleased agents in their own sandbox on specified tasks. Not about the model that gets released to the public.

permalink

fedilink

source

parent

[–] CaptDust@sh.itjust.works 3 points 1 month ago*

At least during the huggingface attacks, the agent sandbox was never intended to have internet access. Openai and independent researchers said the tool broke through that restriction.

permalink

fedilink

source

parent