Australian Politics

1877 readers

161 users here now

A place to discuss Australia Politics.

Rules

This community is run under the rules of aussie.zone.

Recommended and Related Communities

Be sure to check out and subscribe to our related communities on aussie.zone:

Plus other communities for sport and major cities.

https://aussie.zone/communities

founded 3 years ago

MODERATORS

lodion@aussie.zone

Gorgritch_umie_killa@aussie.zone

An OpenAI agent hacked Medicare. Will anyone be held responsible? (theconversation.com)

submitted 13 hours ago by bignose@aussie.zone to c/australianpolitics@aussie.zone

34 comments fedilink hide all child comments

Andrew Cullen:

[…] if a human asks an AI agent to gather health statistics, and the agent hacks a government server to do so, the human could lack the deliberate intent required for a conviction. The AI agent, meanwhile, lacks the legal personhood to be charged, as well as human intentionality.

Current Australia laws effectively treat AI actions as if they are something that just happens to us – like a severe weather event. This shows a glaring loophole in our legal system that does not hold those who make, maintain and use these systems to account when something goes wrong.

you are viewing a single comment's thread
view the rest of the comments

[–] SuspiciousCarrot78@aussie.zone 1 points 5 hours ago (1 children)

Well, in the sense that neither was evidence of runaway AI becoming sapient and deciding to act maliciously, which is often how the “AI hacked X” framing gets presented.

Yes, the Hugging Face incident was more technical / concerning. (For those not in the loop: OpenAI agents in a cyber-security eval sought answers outside the eval environment, found useful material on Hugging Face, discovered HF credentials, and then tried to exploit them. Basically, the agents were trying to cheat on an exam.)

My point was about the narrative around how the media presents “AI hacking”, not that the two incidents were technically equivalent.

[–] Ilandar@lemmy.today 1 points 5 hours ago (1 children)

They didn't hack HuggingFace to cheat on the exam, they had already successfully reverse-engineered the problem by then. The agents hacked HuggingFace because they incorrectly assumed that the scorer would review their transcripts and discover that they had cheated (they were supposed to be isolated and had found a way to communicate with one another without detection). Hacking HuggingFace was an attempt to find a way to obscure/spoof their working out, and that's what makes it so concerning. It had no direct connection to the original task that was set and was a very clear example of agents spiralling out of control in a way that took OpenAI and independent researchers several months to a) become aware of and b) understand correctly.

Here is the METR report on the incident:

https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation

[–] SuspiciousCarrot78@aussie.zone 1 points 4 hours ago* (last edited 4 hours ago) (1 children)

Fair correction. I should have been more specific and not dumbed it down for lurkers / lay audience.

My broader point though is neither that (nor the medicare "hack") is proof that skynet is knocking on the front door.

OTOH, the agents are displaying certain emergent behaviors that are worth mulling over - like the obfuscated back channel communication. Not "sapient" (and I use that word deliberately) but ... curious / problematic, from a control pane level.

[–] Ilandar@lemmy.today 1 points 3 hours ago (1 children)

My broader point though is neither that (nor the medicare “hack”) is proof that skynet is knocking on the front door.

I don't think that's relevant, though. Like this entire sentience/super-intelligence debate that everyone seems so captured by, particularly in the wake of the Amodei letter and renewed calls for regulation and cooperation, is missing the point that the existing models are already being created in a way that prevents the creators from fully understanding what they're creating or how to control it (because of the pace at which they're operating). It's already unsafe and is already causing real world harms.

It really frustrates me that the response to Anthropic, OpenAI and X calling for regulation publicly, or the first high profile security breaches, is corporate conspiracy theories and semantics about how we should classify/characterise the technology. People seem more interested in having their little debate bro moments online than actually getting together and agreeing that we should do something about the real and current problems.

[–] SuspiciousCarrot78@aussie.zone 1 points 2 hours ago* (last edited 2 hours ago)

I do think it's relevant / I think we're arguing the same point from different ends. I'm simply underlining the idea that this whole shooting match isn't AGI and that we should push back on that framing, before we solve the wrong problem in the wrong way.

It's still dangerous, but semantic games of gotcha are not what needs to happen.