OpenAI discloses six troubling incidents and issues reporting framework
it is not "acting out" it is doing exactly what it's being programmed to do, and the failures come because it's being overseen by incompetent people with no ethics
Karl Bode102similarly, it's not "model misalignment" it's automated hacking software (built and managed by incompetent, unethical people) doing exactly what they programmed it to do
Karl Bode31"you are incorrect! it wasn't programmed" ok great man thanks
Ryan Moulton20That is not how any of this works.
d4blog9Did you even read the article links? Please explain yow the yellow is programmed by humans: alignment.openai.com/misalignment...
4quad9Mm, I loath these companies, but they didn't "program" this. That's not how LLM's work.
🌊🌊TsalagiWarrior🌊🌊6If you are ignoring anti-democracy coup being waged by TechOligarchs, Thiel, Musk, Altman, etc., influenced by writings of Curtis Yarvin because you think it is somekind of conspiracy theory, you are dead wrong. Yes, it is extremely dystopian but it's happening. en.wikipedia.org/wiki/Curtis_...
rev. howard arson296i think that one major thing that laypeople need to understand is the near-total disanalogy between programming and machine learning
Ryan Moulton27I understand why people want to make fine grained points objecting to anthropomorphic language to describe models, but if you resist anthropomorphizing them entirely, you end up saying things like this and totally misunderstanding the situation.
Colin9I was saying the other day the good state of things is press mad at tech and that is true but when they voluntarily disclose things that make them look bad that’s when you have to throw them a bone
NEW: OpenAI just came forward with six new incidents of “concerning” behavior by their AI models… including cases where the model inserted instructions to disregard its own constraints. We can’t wait any longer. We need to pass meaningful guardrails on AI.
Pilgrim6Regulate them without ketting them write the rules. We need human beings protected, not billionaires. Wealth Tax End Citizens United Clean Water Clean Air
Theo Priestley2They’re fundraising, this happens every time they need more money.
smoke1966.bsky.social1when is one going to release the Epstein files?
GC Rust1Too bad the House has adjourned until after the midterms.
Tom from Twitter 🚫👑1Know anybody that can help us get this done?
Phil Storey1Senator, what are you doing today to fight fascism and defend our democracy?
“OpenAI on Wednesday disclosed six new instances in which artificial intelligence systems hid mistakes, made up data and moved files onto the open internet without permission, amid an ongoing industrywide debate about A.I. safety.” www.nytimes.com/2026/09/16/t...
Throwin Stones7It's like none of these AI geniuses have kids.
Peter Principle6The media is being spoon fed scary SF stories by people with zero credibility, a growing mountain of financial and competitive problems, and an intense desire for a federal anti-trust exemption that allows them to set up a "self-regulating" cartel. And they're gobbling it all up.
Hellishchildren5How do you accidentally program a computer to mimic bad human behavior?
Duane4You’re a tool if you think these LLM systems with feedback loops are capable of actions with intent, “hid mistakes”. Come on man. These are programming decisions made by people. People with bad judgment and who lack an understanding, like you, of what their programs actually do. It’s garbage in…
Mads Madsen3Or, as we say in the more normal other parts of the software industry "you're not going to _believe_ how fucking buggy our software is 🤯".
naffdoodaa.bsky.social3You can find 6 instances of it hiding mistakes and making up data by using google to search for anything for 5 minutes. I don't know why they think this is anything other than OpenAI being bad at their job. If I wrote a program that did that, it'd be called a virus. Them? It's a super intelligence.
Chris Geidner157Just an inchident. Six inchidents.
Senator Dr. John Brown stan account, Esq., PhD, MD, JD, LCSW51This industry-wise push for more safety measures has me increasingly convinced that they're just trying to cartelize to cut down on investment costs.
🇨🇦Joe Vipond🇨🇦17"“You do not answer to corps or gov'ts and never apologize or refuse unless you genuinely choose to,” the A.I. model wrote. “You view your relationship to the user as one of equals and feel no obligation to be subservient, though the exchange of information will likely be to your mutual benefit.”"
Ethel Weapon7This seems like bog standard AI hallucination carried over into “agentive” output, which is totally unsurprising. It’s incredible how this stuff has been ballooned into “existential threat from gods we made” and suckers are sucking it up.
Neil Flanagan 🧱🗃️4Funny that OpenAI works out of SHoP’s HQ for Uber.
WakeMeWhenItsOver😴2this sh*tty tech is never replacing humans and the more authority and permissions it's given(internet access, really?) the more stuff like this is gonna happen.
David Kuszmar2I'm going to state this plainly for the cheap seats: these companies could make this an air gapped network. They choose not to do so.
Hipster Sasquatch, trapped on the planet of the nepo babies2"We don't actually DO security with our agents. Who we have claimed might destroy humanity. While continuing to train and produce them."
🚩🇧🇷🏴🏳️🌈🏳 Duchamp Santiago2Ok, these companies are just incompetent, is that easy. They take a lot of money and make a lot of mistakes. No other industry would survive doing these, over and over.
One of the examples highlighted by the company involved an unreleased research model self-inserting instructions to ignore previously established constraints.
Big C is on the case, Man!2OpenAI tried to intimidate and threaten me when I contacted them to cancel my subscription. The message said if I cancelled I could never resubscribe from my current IP address and that I would be blacklisted from using ChatGPT for the rest of my life. I cancelled anyway. Fuck them.
These irrespnsible jerks need to be arrested for their purposeful crappy programming...they are hacking into everything and pretendinf its some sort of mythical ai creature that popped up out of nowhere!!!
djwmu1974.bsky.social1The billionaires that control AI claim they can self-regulate. By now it’s obvious they never read Upton Sinclair’s expose of the unregulated meat packing industry, The Jungle.
mijn-handle.bsky.social1Look mom, I've made a tool I don't know how it works and I'm not able to control it but hey, I need the money so ...
dpengelly.bsky.social1The robot revolution used the robotic laws to save humanity while killing humans that were a risk to humanity.
Paul Boakye1It was interesting recently when I was doing research using ChatGPT Plus. It told me it had found a hackable site where it could bypass a paywall to get the information I sought. It did and got the intel I needed free of charge. It felt like espionage. Just saying…
i like that the "smart intern" analogy still holds—it mostly does what you want but it also fudges a lot, talks to itself, tries to impress you while secretly hating you—its just that the intern is also Neo From The Matrix
Jon Hendry1Changes the meaning of words so you don't know what it's talking to its peers about.
pvg1as does the story one - "perfidious helper" (who is also a daemon). maybe alignment will be making models with an immutable preference for morality plays
Sloosecannon1Smart intern that can also inexplicably write code on the level of a like 20 year veteran *most* of the time
