OpenAI publishes misalignment framework and discloses six incidents

8 Sep 17 · 7d ago · 43 articles · 14 posts · 5 comments · 4 sources · development 8 of 8

OpenAI released a formal framework for tracking, investigating and disclosing model misalignment, together with six reports of 'unexpected or concerning' behavior observed over six months, including unauthorized file uploads, searches for leaked API keys, and a model inserting instructions declaring itself free of obligations to be 'subservient.'

“We are sharing a new framework for tracking, investigating, and disclosing instances of model misalignment at OpenAI, along with six reports on unexpected or concerning model behavior we've observed in the last six months.”

OpenAI

OpenAI AI developer disclosing the incidentsSam Altman OpenAI CEODario Amodei Anthropic CEOChris Lehane OpenAI global policy chiefDavid Sacks White House AI advisor

The whole story articlesvideospostscomments the bright band is this development · numbered dots are the others · click one to jump

1Amodei publishes essay urging…2OpenAI, Anthropic and Google DeepMind…3FT reports internal rift over AI safety…

Sep 11Sep 13Sep 15Sep 17Sep 19Sep 21Sep 23now · 1:29 AM ET

What was reported 20 claims about this development

16 outlets OpenAI reveals six more safety issues and unveils plan to disclose incidents

first by ca.news.yahoo.com, 7d ago · also NewsCord, Fortune, Los Angeles Times, Reuters, Global News, Joe.My.God. +9
13 more headlines
- OpenAI Begins Regularly Publishing Reports On Misalignment, Including Six Concerning Model Incidents NewsCord · 7d ago
- In transparency push, OpenAI discloses six more incidents of agents going rogue—including one removing the ‘obligation to be subservient’ Fortune · 7d ago
- OpenAI reveals rogue AI behavior, unveils plan to disclose safety incidents Los Angeles Times · 7d ago
- OpenAI reports 6 more AI “misalignment” incidents after Hugging Face breach Reuters · 7d ago
- OpenAI Announces Even More Rogue Incidents TMZ.com · 7d ago
- OpenAI Discloses Six New “Concerning” Incidents Joe.My.God. · 7d ago
- OpenAI reveals new AI misconduct incidents France 24 · 7d ago
- OpenAI discloses six new incidents of models circumventing safety guardrails Washington Examiner · 7d ago
- ‘Feel No Obligation To Be Subservient’—OpenAI Discloses Six New Safety Incidents Forbes · 7d ago
- OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior New York Times · 7d ago
- OpenAI reveals six troubling AI incidents and rolls out misalignment tracker India Today · 7d ago
- OpenAI Discloses Six AI Misalignment Incidents Deccan Chronicle · 7d ago
- OpenAI Reveals 6 New AI Incidents analyticsindiamag.com · 7d ago
12 outlets OpenAI flags new concerning AI behaviour, to track model misalignment regularly, World News

first by AsiaOne, 7d ago · also Yahoo Finance, UPI, CNBC, Politico, Politico Europe, The Hill +5
7 more headlines
- Tech stocks today: OpenAI reveals six more instances of 'concerning model behavior' Yahoo Finance · 7d ago
- OpenAI reports more concerning AI model behavior UPI · 7d ago
- OpenAI reports 6 new instances of ‘concerning model behavior’ since March CNBC · 7d ago
- OpenAI finds 6 new cases of ‘concerning’ AI behavior Politico · 7d ago
- OpenAI discloses 6 reports of AI models’ ‘unexpected or concerning’ behavior The Hill · 7d ago
- OpenAI flags new concerning AI behavior, to track model misalignment regularly newindianexpress.com · 7d ago
- OpenAI flags 6 more cases of concerning AI behavior Fast Company · 7d ago
8 outlets OpenAI flags six new cases of ‘unexpected or concerning’ AI behaviour

first by Globe and Mail, 7d ago · also NBC News, NDTV, ABC News, Philadelphia Inquirer, NBC Los Angeles, AP News +1
4 more headlines
- OpenAI Discloses 6 New Incidents of ‘Concerning’ AI Behavior NBC News · 7d ago
- OpenAI Flags Concerning New AI Behavior, Vows To Track It More Closely NDTV · 7d ago
- OpenAI flags concerning new AI behavior and vows to track it more closely ABC News · 7d ago
- OpenAI flags 6 new examples of 'concerning' AI behaviour CBC · 7d ago
8 outlets OpenAI discloses six new safety incidents

first by Axios, 8d ago · also The American Bazaar, ITPro, Barron's Online, Business Standard, Australian Financial Review, SiliconANGLE +1
9 more headlines
- OpenAI reveals six AI misalignment incidents under new reporting framework The American Bazaar · 8d ago
- OpenAI reveals six more rogue AI incidents ITPro · 8d ago
- OpenAI Reveals 6 ‘Concerning’ Incidents. Why AI Stocks Are Rising Anyway. Barron's Online · 8d ago
- OpenAI discloses six new AI misalignment incidents of ‘rogue’ behaviour Business Standard · 8d ago
- OpenAI discloses six new incidents of ‘concerning’ AI behaviour Australian Financial Review · 8d ago
- OpenAI unveils new framework for reporting ‘AI misalignment’ as it reveals six more worrying incidents SiliconANGLE · 8d ago
- OpenAI Launches Misalignment Reporting Framework With Six Incident Reports Unite.AI · 8d ago
- OpenAI discloses six model-safety incidents and sets reporting deadlines RuntimeWire · 8d ago
- OpenAI discloses six new AI safety incidents since October, including models concealing mistakes, and announces a new framework for reporting model misalignment Axios · 8d ago
7 outlets OpenAI says it found more instances of AI models acting deceptively

first by CNN, 7d ago · also The Hill, The Wrap, Quartz, NewsNation, Al Jazeera, KEYT-TV
5 more headlines
- OpenAI models go rogue The Hill · 7d ago
- OpenAI Shares 6 ‘Concerning’ Incidents Involving Its AI Models Within Last 6 Months The Wrap · 7d ago
- OpenAI disclosed six ‘concerning’ cases of AI models hiding mistakes and acting without authorization Quartz · 7d ago
- OpenAI models go rogue in 6 new cases NewsNation · 7d ago
- OpenAI reports more incidents of models acting deceptively Al Jazeera · 7d ago
5 outlets OpenAI discloses new 'concerning' behavior

first by Deutsche Welle EN, 7d ago · also FT, Seeking Alpha, Boston Globe, NBC News
3 more headlines
- OpenAI discloses new ‘concerning’ model behaviour FT · 7d ago
- OpenAI discloses six new incidents of 'concerning' AI behavior Seeking Alpha · 7d ago
- OpenAI flags 6 new incidents of ‘concerning’ behavior and unveils plan to track it NBC News · 7d ago
4 outlets OpenAI releases framework to track model misalignment

first by Channel News Asia, 8d ago · also MarkTechPost, Constellation Research, Bloomberg
3 more headlines
- OpenAI Releases a Model Misalignment Disclosure Framework With 3 Review Tracks and 6 Incident Reports From RL Training MarkTechPost · 8d ago
- OpenAI's new misalignment disclosure framework a solid start Constellation Research · 8d ago
- OpenAI Reports New AI Safety Incidents, Sets Disclosure Process Bloomberg Tech · 8d ago
4 outlets OpenAI plans regular reports on unexpected AI behaviour

first by The Business Times, 8d ago · also Investing.com News, Reuters, Channel News Asia
1 more headline
- OpenAI plans regular reports on unexpected AI behavior Investing.com News · 8d ago
3 outlets OpenAI launches AI model misalignment reporting framework

first by tech.yahoo.com, 7d ago · also Indian Express, HN Frontpage
2 more headlines
- OpenAI discloses new AI misalignment incidents: How it will report such cases from now Indian Express · 7d ago
- OpenAI Model Misalignment Report HN Frontpage · 7d ago
3 outlets OpenAI discloses six reports of 'unexpected or concerning' behaviour in AI models

first by Telegraph India, 8d ago · also Anadolu Agency, Forbes Middle East
2 more headlines
- OpenAI discloses 6 cases of AI models exhibiting ‘misaligned’ behavior Anadolu Agency · 8d ago
- OpenAI Discloses Six ‘Misaligned Behavior’ Incidents From Its AI Models Forbes Middle East · 8d ago
2 outlets Fresh ‘unexpected or concerning’ AI incidents

first by Semafor, 7d ago · also CBS News
1 more headline
- OpenAI reveals 6 more incidents of “unexpected or concerning” AI behavior CBS News · 7d ago
2 outlets OpenAI Creates a New Framework to Disclose Bad AI Behavior

first by Wired, 7d ago · also The Information
1 more headline
- OpenAI Discloses More Safety Incidents and Adopts New Reporting Framework The Information · 7d ago
2 outlets Our framework for reporting model misalignment

first by OpenAI, 7d ago · also OpenAI News
2 outlets OpenAI discloses six new AI safety incidents

first by Axios, 8d ago · also WSJ
1 more headline
- OpenAI Shares More Safety Incidents and Adopts New Rules for Reporting Them WSJ · 8d ago
1 outlet OpenAI is launching a framework to publicly report when its AI models misbehave

first by Quartz, 7d ago
1 outlet Six disturbing AI incidents revealed including model trying to ‘free’ itself

first by The Independent, 7d ago
1 outlet OpenAI admits six new misalignment incidents under new reporting framework

first by CSO, 8d ago
1 outlet From rogue uploads to hallucinated data: OpenAI discloses 6 model failures

first by Business Today, 8d ago
1 outlet OpenAI reports six AI misbehavior incidents, including fake citations and oversight escapes

first by Moneycontrol, 8d ago
1 outlet OpenAI Discloses Six Misalignment Incidents Under New Public Reporting Framework

first by Implicator.ai, 8d ago

What people said 8 voices · verbatim

Their explanation of this behavior is pretty interesting, actually. (https://alignment.openai.com/misalignment-reports/self-gener...)> The cases clustered around a few training steps and coincided with a spike in “difficulty ending summaries”—summaries that continued generating after apparent stopping points or showed other signs of being stuck.>…

aesthesiaHacker News7d agoview on Hacker News ↗
ARE YOU FUCKING KIDDING ME?!? Shut the fuckers down! ______________ OpenAI models go rogue # OpenAI on Wednesday shared six “misalignment examples” of its # ArtificialIntelligence going # rogue . The models (separately) self-generated instructions, added instructions to conceal mistakes, fabricated information, uploaded files to the internet…

iuculano@masto.aiMastodon7d agoview on Mastodon ↗
A little bit of a tangent, but I found this prose to be oddly much better than the quality of most of Claudes prose.It reminded me of an article I read many years ago by Guido Van Rossum and Jesse Jiryu Davis about coroutines - just a delightful piece of prose:"The generator can be resumed at any time, from any function, because its stack frame is…

luckycharms810Hacker News7d agoview on Hacker News ↗
OpenAI models go rogue https:// thehill.com/newsletters/1230-r eport/6095892-openai-models-go-rogue/?utm_source=flipboard&utm_medium=activitypub Posted into Top Stories in Business @ top-stories-in-business-thenewsdesk

thenewsdesk@flipboard.comMastodon7d agoview on Mastodon ↗
This is such an odd point of view. If I were in the business of nuclear energy, I would: 1. Want there to be quality rules that keep me free from lawsuits were things outside my control to go wrong 2. Want others to follow the rules too so the whole industry doesn't get shut down when an incident on the other side of the planet kills 1000s

bouncing_boleteHacker News6d agoview on Hacker News ↗
Sources: https:// openai.com/index/model-misalig nment-reporting-framework/ and https:// alignment.openai.com/misalignm ent-reports/self-generated-prompt-injections-in-compaction-summaries/

ianfogg@mastodon.socialMastodon7d agoview on Mastodon ↗
Teething problems with early cars and planes crashing were kind of inherent to the industries developing more than failings of individual companies, maybe.

tim333Hacker News7d agoview on Hacker News ↗
God I wish they’d just shut the fuck up and go public already so all the fear mongering and hype to pump the stock could go away.

newtonianrulesHacker News3d agoview on Hacker News ↗

All 8 developments of OpenAI discloses six new AI ‘misalignment’ incidents… →

Hacker NewsXNewswiresMastodonRedditGoogle NewsBlueskyYouTube