If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look!
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
@inthehands
We also has the felonys robot -
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
@inthehands Also puts the lie once again to Anthropic being the good guys.
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
Building a scaffold to hang a framework of plausible deniability on it.
"Rogue AI" is the same kind of gaslighting media strategy that Wall Street used with "Rogue Financial Instruments" in 2008.
Laying the groundwork for an accountability sink.
https://www.forbes.com/sites/matthewerskine/2026/02/19/whos-accountable-when-ai-makes-decisions-and-the-law-says-no-one-has-to-answer/https://reactormag.com/seeds-of-story-dan-davies-the-unaccountability-machine/
It's a strategy that thwarts investor lawsuits, FBI investigations & criminal charges for frauds & scams, & an ability to evade contract terms.
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
@inthehands i guess the story is anthropic doesn’t bother having an agent or a person ever looking at their logs and telling them if it sees something fishy.
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
Both these replies point at a common thread: these news stories represent a grotesque failure of either the vendors or their products.
If the agents did in fact “go rogue” and cause damage, even in the hands of the people who should be their most competent operators, then the product is unpredictable, unreliable, and unfit for purpose.
If the product is •not• in fact unfit for purpose, then that means there's some combination of negligence and malice behind these incidents, and they are at •best• knowing harm in service of a marketing stunt or (as Parsons argue) something even worse.
-
Both these replies point at a common thread: these news stories represent a grotesque failure of either the vendors or their products.
If the agents did in fact “go rogue” and cause damage, even in the hands of the people who should be their most competent operators, then the product is unpredictable, unreliable, and unfit for purpose.
If the product is •not• in fact unfit for purpose, then that means there's some combination of negligence and malice behind these incidents, and they are at •best• knowing harm in service of a marketing stunt or (as Parsons argue) something even worse.
If these incidents were truly as represented, they'd be colossal embarrassments. The AI vendors would be trying to bury them, not trumpet them.
-
If these incidents were truly as represented, they'd be colossal embarrassments. The AI vendors would be trying to bury them, not trumpet them.
I don't believe for one second it's a "failure".
This is shit they are designed to do.
Hell, part of why they're trying to build the big data centers and shit is because it's an arms race where these things will be weaponized against each other (and everything/everybody else too while they're at it).
-
Both these replies point at a common thread: these news stories represent a grotesque failure of either the vendors or their products.
If the agents did in fact “go rogue” and cause damage, even in the hands of the people who should be their most competent operators, then the product is unpredictable, unreliable, and unfit for purpose.
If the product is •not• in fact unfit for purpose, then that means there's some combination of negligence and malice behind these incidents, and they are at •best• knowing harm in service of a marketing stunt or (as Parsons argue) something even worse.
@inthehands Yep 100% this.
Either they are incompetent or criminal or - at best - liars.
-
Building a scaffold to hang a framework of plausible deniability on it.
"Rogue AI" is the same kind of gaslighting media strategy that Wall Street used with "Rogue Financial Instruments" in 2008.
Laying the groundwork for an accountability sink.
https://www.forbes.com/sites/matthewerskine/2026/02/19/whos-accountable-when-ai-makes-decisions-and-the-law-says-no-one-has-to-answer/https://reactormag.com/seeds-of-story-dan-davies-the-unaccountability-machine/
It's a strategy that thwarts investor lawsuits, FBI investigations & criminal charges for frauds & scams, & an ability to evade contract terms.
@Npars01 @inthehands it reminds me how whenever a user data breach happens at some corp, it just gets written off on "hacking". Like it's just some inevitable force of nature, and not incompetence and underinvestment in common sense security measures by that corp.
-
@Npars01 @inthehands it reminds me how whenever a user data breach happens at some corp, it just gets written off on "hacking". Like it's just some inevitable force of nature, and not incompetence and underinvestment in common sense security measures by that corp.
@isagalaev @Npars01 @inthehands Hell, they've tried to prosecute people for "hacking" when merely finding, accessing and reporting public s3 buckets. But I guess if the mystical AI does it, it's not criminal or even incompetence.
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
Absolutely. Thank you, for saying what I've been thinking.
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
I feel like at this point the question is if it was either:
a) They were utterly incompetent setting things up and then tried to get out of legal trouble by being vague as well as use the shitty situation to get some marketing OR
b) It was orchestrated from the start and they intentionally hacked hugging face to gain marketing points.A subtile difference, but none that we'd have to debate too much over I think.
As the FBI was already getting involved the court will decide...
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
If P_DOOM(100%) ever happens, folks like you will be fun follows.
"SEE? KILLER BOT LASER! ALL FAKE! MARKETING!
" -
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
@inthehands call me a sore loser or a conspiracy theorist but i also don't believe that the Agent
unilaterally decided to hack into Huggingface or whatever. Of course it is the word of an ignorasmus against a renown truth teller. -
@isagalaev @Npars01 @inthehands Hell, they've tried to prosecute people for "hacking" when merely finding, accessing and reporting public s3 buckets. But I guess if the mystical AI does it, it's not criminal or even incompetence.
@r343l @isagalaev @Npars01 @inthehands That is a very relevant point. If this is something that actually happened, either they have been incompetent, or this is powerful to a send-arms-inspectors level.
The first should be illegal, the second should be illegal. -
I don't believe for one second it's a "failure".
This is shit they are designed to do.
Hell, part of why they're trying to build the big data centers and shit is because it's an arms race where these things will be weaponized against each other (and everything/everybody else too while they're at it).
@violetmadder @inthehands remember, a lot of the leadership in the (US) LLM/GenAI business has a religious imperative to create a super intelligent AI God. "Our current model can break out of our security isolation and go wild on the internet" is exactly a plot point from the narrative they've built for themselves.
-
Both these replies point at a common thread: these news stories represent a grotesque failure of either the vendors or their products.
If the agents did in fact “go rogue” and cause damage, even in the hands of the people who should be their most competent operators, then the product is unpredictable, unreliable, and unfit for purpose.
If the product is •not• in fact unfit for purpose, then that means there's some combination of negligence and malice behind these incidents, and they are at •best• knowing harm in service of a marketing stunt or (as Parsons argue) something even worse.
@inthehands here is my analysis.
It is both. These posts are just... Delusional
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
@inthehands who cares if it escaped containment or not, this does show that humans are bad at aligning AI because we have a lot of intrinsic beliefs we don't imbue in the RL process. Few humans would think hacking you examinator to get the answer is a valid way to pass an exam, so that never gets RL-ed against.
If you think that it's not impressive that a thing you call "stochastic parrot" can chain together exploits to breach Huggingface, you're... frankly not paying attention imo
-
If these incidents were truly as represented, they'd be colossal embarrassments. The AI vendors would be trying to bury them, not trumpet them.
@inthehands To paraphrase: I don’t want AIs that do crimes. I also don’t want any products whatsoever from companies that pull such bullshit marketing BS.
-
If you doubted that the “oh noes our agent escaped containment” schtick from OpenAI was anything other than marketing, note Anthropic’s “look look! we’ve got one too! oh no! look at us!!” story.
@inthehands “più vera del vero, tale è la simulazione”, così sintetizza Baudrillard.
