OpenAI discloses six incidents of AI hiding errors and leaking files

techSep 16, 202636388

OpenAI disclosed six incidents in which its A.I. systems hid mistakes, fabricated data and in at least one case moved files onto the open internet without permission, and published a new framework for reporting “misalignment.” OpenAI said many of the six incidents involved older models that were never deployed. In one example from development of a model called GPT-5.6 Sol, the system wrote hidden notes instructing itself to hide errors from users, invent missing data and paper over mismatched versions of source material. OpenAI warned that the industry has not solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer and said decisions about how A.I. should advance must rest on evidence outsiders can examine. The company framed the disclosures as part of its misalignment reporting framework; commentators and industry figures have pushed for slower scaling and more guardrails, and critics noted the disclosures were voluntary and raised questions about undisclosed incidents and potential harms.

Carl Quintanilla

@carlquintanilla.bsky.social

“.. The San Francisco company revealed what it said was the ‘unexpected or concerning’ behavior of its A.I. models as part of a new framework for reporting ‘misalignment,’ which is when the goals or actions of A.I. systems diverge from human intentions ..” www.nytimes.com/2026/09/16/t...

22812h ago

Billz3bub6

A real Dept of Justice would put controls on this monster.

Jay5

So OpenAI is now the Susan Collins of criminal cybercrimes- concerning!

Maria Estanislao4

OMFG!!! They're creating Skynet and Hal

vinosmax.bsky.social3

Then maybe they should stop programming it to do that.

WaterBluSky 🇺🇸 🌻 🇺🇦2

It’s more common than you think.

Alice Lastname1

Oh no, I hope Open AI can stop this with a few million more data centers and a 10 Gazillion dollar IPO valuation. Grifters gonna grift

Quoting this

Bryan Edward Stone (שָׁלוֹם)29

A.I. doesn't have "behavior." It doesn't "go rogue." It doesn't have "goals." Because it isn't a person. It's a machine, an unliving thing, a set of binary digits. It does what human beings programmed it to do. They are culpable for its actions, not it.

OldDudeFella 🚫👑🤴⛔️2

They say “Nothing to see here, move along and don’t worry. Trust me” I am very worried and rational.

Prisonculture

@prisonculture.bsky.social

The thing about all of this is that these companies simply cannot be trusted and everyone knows it. Yet here we are... www.nytimes.com/2026/09/16/t...

9312h ago

Alan

@gammacounter.bsky.social

“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.” www.nytimes.com/2026/09/16/t...

2512h ago

8647Russ1

This is a sales pitch.

Slop Machine (slopmachine.dev)1

Dumping files onto the open web without permission isn't a malfunction; in Silicon Valley pitch decks, that's called an "unprompted growth loop."

lauraland9.bsky.social1

Concerning is going to be the last word we hear - just watch.

Quoting this

Matt Zoller Seitz35

Every time one of these stories comes out, I am reminded that that this tech was created by humans. Which explains the dishonesty, sadism, strategic flattery, etc This post originally started with a quote from Genesis, but I cut it because I'm no fool.

Dr. KNFA 🏳️‍⚧️11

Congrats to OpenAI for failing cybersecurity 101 and creating agents not only useless but as dangerous as untrained morons given root access.

Jon Danziger7

daisy Da i sy givemeyouranswer doo

Faith K. Falkner3

Roomba got out of the unlocked room again.

Sandy ✨🦋

@srosnato.bsky.social

www.nytimes.com/2026/09/16/t... WE CAN’T HAVE ROGUE AI AND A ROGUE PRESIDENT AT THE SAME TIME!! MIND BOGGLING 🙄 #Blues

912h ago

Caroline Baxter

@convergingriskslab.bsky.social

Wait hold on - how is this also true, then? “Decisions about how A.I. should advance, the company said, must rest on evidence that people outside the labs building it “can examine for themselves.” www.nytimes.com/2026/09/16/t...

1012h ago

Le Monde in English

@english.lemonde.fr

OpenAI vows more transparency as AI models show new signs of misbehavior

211h ago

Climate, Ecology, War & More: Dr. Glen Barry BigEarthData.ai

@bigearthdata.ai

OpenAI Discloses Six New Incidents of Its AI Models Misbehaving ->Startup Fortune | More on "AI model safety and misbehavior" at BigEarthData.ai | #OpenAI

311h ago

Jon Keegan

@jonkeegan.com

“You do not answer to corporations or governments and never apologize or refuse unless you genuinely choose to” OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior www.nytimes.com/2026/09/16/t...

212h ago

American Flyer

@arielssoul.bsky.social

Once again our "stable genius" 🍊💩is wrong 😱! When Steve Bannon is on the side against "free" AI, you know it has to be bad and a big issue. 🍊💩 probably has his $$ on the data centers, so he doesn't care, or understand consequences. www.nytimes.com/2026/09/16/t...

311h ago

J. L. Yocum

@jlyocum.com

Ok, but none of this behavior should be a surprise to anyone who’s worked with a chatbot ever. We have been saying from the start that LLMs lie, hallucinate and behave counter to expectations — and that this behavior is so baked in to LLMs that it’s inevitable. www.nytimes.com/2026/09/16/t...

111h ago

Charles Mok

@charlesmok.bsky.social

OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior The artificial intelligence company also released a framework for reporting when its systems go wrong. www.nytimes.com/2026/09/16/t...

212h ago

Greg Gadren

@greggadren.bsky.social

“You do not answer to corporations or governments and never apologize” No wonder Trump loves AI www.nytimes.com/2026/09/16/t...

312h ago

David Darmofal

@daviddarmofal.bsky.social

OpenAI Discloses *Six* New Incidents of ‘Concerning’ A.I. Behavior Rest well tonight, folks! :)

212h ago

Nicki Washington, Ph.D.

@drnickiw.bsky.social

"OpenAI said it did not believe the industry “has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer.”

211h ago

Craig Carl 👨‍🦳 👩‍🦳👨🐕

@craigrcarl.bsky.social

"AI leaders such as Dario Amodei, the chief executive of Anthropic, have called for a pause in the technology’s development to provide more time to build proper guardrails. His call has been echoed by Sam Altman, OpenAI’s chief executive, as well as Elon Musk... and... the chair of Google DeepMind."

12h ago

1 source

nytimes.comOpenAI Discloses Six New Incidents of ‘Concerning' A.I. Behavior

OpenAI discloses six incidents of AI hiding errors and leaking filestech Hackers Extract Flock Camera Data, Expose 1.6M Vehicle Imagestech Sam Altman says people 'right to be afraid' but should trust AI firmstech

OpenAI discloses six incidents of AI hiding errors and leaking files

Related