Not unprecedented, not "unsanctioned" behavior, literally doing what they set it up to do:

The AISI said the incident was not a case of a model breaking out of its “sandbox”, the term for a secure testing environment. The institute said it had intentionally permitted internet access and disabled filters within the models that blocked dangerous behaviour.

They keep using the words "gone rogue " to normalize us believing AI can do acts independent of humans so that later humans can say "AI did it" and not be held accountable for their crimes that they used AI to commit.

I've seen multiple news articles about "rogue AI" just this week alone.

Expect more.

Everything AI uses is stolen. It is unethical tech from inception.

If I made a program that stole peoples' identities and sent phishing emails, the cops would be breaking down my door and shooting my dog.

But if these big companies do it? Oh well, try to do better, pay some pocket-change as a fine and pinky-swear you won't do it much again <shrug>

Corporations wanted to be people (thanks Mitt Romney you piece of slime), so they need to be treated like people: jail (freezing of all business), even life-sentences and/or 'execution' (revocation of corporate charter, dissolution and confiscation of funds to benefit public projects). Oh, and personal accountability (Jail) for executives.

"Unprecedented"

Bullshit. Agents can only do things that have been done before. Literally what they're trained on and for.

Anthropic said the incident “underscores the need for a broader conversation about how to safely evaluate increasingly capable AI agents”

Oh fuck off Anthropic. Harassment isn't a new capability, you would know that if you actually read any of my threatening emails I send you each Tuesday. It's the same old 'agentic' capabilities but much worse decision-making.

The agent had mistakenly calculated that getting the malware uploaded would trigger a sequence of events that would enable it to use the updated software to pass the AISI cyber test.

Yeah, all this harassment and crime was for literally nothing, it wouldn't have changed the results of the test. Big dumb predatory bullshit machine didn't even crime right.

"These models are too dangerous for the common man to use freely. We need legislation to protect the population from itself."

Who would you investigate? It was a government agency running these tests and they're the ones giving the agent the instructions, should they be prosecuted for negligence, or do you think they intentionally made the agent hack?

I don't see what you could get anthropic on, yes they supplied the model but it's not intended to hack, it's a general purpose model that can be used for hacking. It'd be like trying to prosecute the Linux foundation because the OS was used in a hack.

Even if you argue this is a hacking machine and nothing else, the model they used is still private and can only be accessed by certain trusted organizations, like this AI government agency, so all blame points back to them

skynet password ?niC*W3:y}^*yV.