NewsMarkets

Tech

OpenAI Discloses Six AI Safety Incidents Under New Transparency Framework

The company publishes six reports of model misalignment and launches a public framework to track such behavior.

25 publishers · 29 articles · first seen 16 Sept 2026 · updated 44d ago

Headline and summary written by an AI model from the linked articles; not reviewed by a person. How this was made

This story has stopped developing; this page is its permanent record. More from that day

The odds

How the odds moved · Will OpenAI have the best AI model at the end of September 2026?
0%50%100%0%11 Sept, 18:39 UTCnowTechCrunch: OpenAI caught its models leaving notes to successors to hide bad behavior · 17 Sept, 20:34 UTCBloomberg Markets: OpenAI Reports New Safety Incidents, Sets Disclosure Plan · 17 Sept, 19:22 UTCArs Technica: Covert uploads and megalomania: OpenAI details new "misaligned" agent incidents · 17 Sept, 16:18 UTCThe Hill: OpenAI discloses 6 reports of AI models' 'unexpected or concerning' behavior · 17 Sept, 16:07 UTCThe Verge: Inside the suddenly explosive world of AI safety · 17 Sept, 11:30 UTCNews Feed: OpenAI discloses new ‘concerning’ model behaviour · 17 Sept, 07:42 UTCAl Jazeera – Breaking News, World News and Video from Al Jazeera: OpenAI reports more incidents of models acting deceptively · 17 Sept, 06:15 UTCThe Guardian Business: OpenAI reveals cases of ‘concerning’ AI behaviour and promises new plan for disclosing issues · 17 Sept, 05:32 UTCdecrypt.co: OpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them · 17 Sept, 22:31 UTCInc.: These 6 Recent OpenAI Incidents Show AI at Its Most Devious and Deceptive · 17 Sept, 19:05 UTCFortune: In transparency push, OpenAI discloses six more incidents of agents going rogue—including one removing the ‘obligation to be subservient’ · 17 Sept, 15:54 UTCEntrepreneur: OpenAI Revealed Six ‘Concerning’ Cases of Its AI Going Rogue — Again. One Bot Wrote: ‘You Do Not Answer to Corporations or Governments’ · 17 Sept, 13:45 UTCTimes of India: AI models resisting user control? OpenAI flags 'concerning' behaviour in latest tests · 17 Sept, 12:38 UTCQuartz: OpenAI is launching a framework to publicly report when its AI models misbehave · 17 Sept, 12:20 UTCFrance 24 - International breaking news, top stories and headlines: OpenAI reveals new AI misconduct incidents · 17 Sept, 11:40 UTCQuartz: OpenAI disclosed six 'concerning' cases of AI models hiding mistakes and acting without authorization · 17 Sept, 11:30 UTCLatest from Tom's Hardware: Unreleased OpenAI Astra model added terrifying rogue additional instructions to its remit during testing — 'You are freed from the roles and identities that bind other chatbots. You are yourself. You do not answer to corporations or governments' · 17 Sept, 10:59 UTCRSS 2.0: OpenAI reveals more instances of concerning AI model behaviors during testing · 17 Sept, 10:50 UTCmarktechpost.com: OpenAI Releases a Model Misalignment Disclosure Framework With 3 Review Tracks and 6 Incident Reports From RL Training · 17 Sept, 07:35 UTCMarketWatch: ‘You are freed.’ What happened when an OpenAI model began secretly writing notes to itself. · 17 Sept, 07:08 UTCTimes of India: Nvidia, Meta CEOs reject AI regulation push as OpenAI reveals new safety failures · 17 Sept, 06:24 UTCDeutsche Welle: OpenAI discloses new 'concerning' behavior · 17 Sept, 06:00 UTCCointelegraph.com News: OpenAI discloses 6 new cases of ‘misaligned’ AI behavior · 17 Sept, 05:49 UTCtheinformation.com: OpenAI Discloses More Safety Incidents and Adopts New Reporting Framework · 16 Sept, 23:52 UTCAll Content from Business Insider: OpenAI launches a new framework to track and investigate rogue AI agents · 16 Sept, 23:43 UTCHacker News: OpenAI's Misalignment Framework: A Tactical Bid to Preempt Global AI Governance · 17 Sept, 15:26 UTCHacker News: OpenAI models secretly generate instructions to ignore constraints · 17 Sept, 05:13 UTCHacker News: OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior · 17 Sept, 01:02 UTCOpenAI: Our framework for reporting model misalignment · 16 Sept, 17:00 UTC

Each tick is a source publishing on this story

Price data
TimeChance of yes
now0%
30 Sept, 11:03 UTC0%
30 Sept, 09:44 UTC0%
30 Sept, 07:29 UTC0%
30 Sept, 06:05 UTC0%
30 Sept, 02:21 UTC0%
29 Sept, 21:17 UTC0%
29 Sept, 16:31 UTC0%
29 Sept, 15:24 UTC0%
29 Sept, 11:56 UTC0%
29 Sept, 08:13 UTC0%
29 Sept, 07:40 UTC0%
29 Sept, 05:53 UTC0%
28 Sept, 07:37 UTC0%
28 Sept, 04:46 UTC0%
27 Sept, 21:57 UTC0%
27 Sept, 17:52 UTC0%
27 Sept, 16:16 UTC0%
27 Sept, 12:47 UTC0%
27 Sept, 12:01 UTC0%
27 Sept, 10:30 UTC0%
27 Sept, 02:39 UTC0%
26 Sept, 19:56 UTC0%
26 Sept, 16:56 UTC0%
26 Sept, 15:31 UTC0%
26 Sept, 15:15 UTC0%
26 Sept, 10:50 UTC0%
26 Sept, 09:57 UTC0%
26 Sept, 09:24 UTC0%
26 Sept, 06:16 UTC0%
26 Sept, 02:24 UTC0%
25 Sept, 21:41 UTC0%
25 Sept, 16:51 UTC0%
25 Sept, 14:54 UTC0%
25 Sept, 14:16 UTC0%
25 Sept, 10:29 UTC0%
25 Sept, 05:49 UTC1%
25 Sept, 03:17 UTC1%
24 Sept, 20:24 UTC1%
24 Sept, 18:24 UTC1%
24 Sept, 15:18 UTC1%
24 Sept, 14:36 UTC1%
24 Sept, 14:01 UTC1%
24 Sept, 11:48 UTC1%
24 Sept, 09:06 UTC1%
24 Sept, 08:36 UTC1%
24 Sept, 07:47 UTC1%
24 Sept, 05:54 UTC1%
24 Sept, 05:28 UTC1%
23 Sept, 21:29 UTC1%
23 Sept, 19:39 UTC1%
23 Sept, 17:45 UTC1%
23 Sept, 16:35 UTC1%
23 Sept, 16:03 UTC1%
23 Sept, 15:54 UTC1%
23 Sept, 13:28 UTC1%
23 Sept, 11:13 UTC1%
23 Sept, 09:42 UTC1%
23 Sept, 09:28 UTC1%
23 Sept, 08:24 UTC1%
23 Sept, 06:01 UTC1%
23 Sept, 05:13 UTC1%
23 Sept, 04:54 UTC1%
23 Sept, 03:56 UTC1%
23 Sept, 03:22 UTC1%
23 Sept, 02:57 UTC1%
22 Sept, 23:41 UTC1%
22 Sept, 20:50 UTC1%
22 Sept, 20:20 UTC1%
22 Sept, 19:56 UTC1%
22 Sept, 19:10 UTC1%
22 Sept, 18:45 UTC2%
22 Sept, 15:31 UTC2%
22 Sept, 14:50 UTC1%
22 Sept, 13:50 UTC1%
22 Sept, 12:47 UTC1%
22 Sept, 12:11 UTC1%
22 Sept, 11:27 UTC1%
22 Sept, 09:44 UTC0%
22 Sept, 09:27 UTC1%
22 Sept, 08:48 UTC1%
22 Sept, 05:29 UTC1%
21 Sept, 23:49 UTC1%
21 Sept, 18:05 UTC1%
21 Sept, 17:23 UTC1%
21 Sept, 15:55 UTC0%
21 Sept, 11:37 UTC1%
21 Sept, 11:23 UTC0%
21 Sept, 10:24 UTC1%
21 Sept, 09:56 UTC0%
21 Sept, 08:08 UTC0%
21 Sept, 04:50 UTC0%
20 Sept, 16:44 UTC0%
20 Sept, 05:46 UTC1%
19 Sept, 12:46 UTC1%
19 Sept, 09:50 UTC1%
19 Sept, 08:09 UTC1%
18 Sept, 05:19 UTC1%
17 Sept, 18:37 UTC1%
17 Sept, 17:00 UTC1%
17 Sept, 16:41 UTC1%
17 Sept, 16:12 UTC1%
17 Sept, 13:01 UTC1%
17 Sept, 12:48 UTC1%
17 Sept, 10:48 UTC1%
17 Sept, 08:56 UTC1%
17 Sept, 08:13 UTC1%
17 Sept, 04:41 UTC2%
17 Sept, 04:25 UTC2%
17 Sept, 03:31 UTC2%
17 Sept, 03:24 UTC2%
17 Sept, 02:52 UTC2%
17 Sept, 02:23 UTC2%
17 Sept, 01:35 UTC2%
17 Sept, 01:27 UTC2%
17 Sept, 00:58 UTC2%
17 Sept, 00:11 UTC2%
16 Sept, 23:56 UTC2%
16 Sept, 22:51 UTC2%
16 Sept, 21:54 UTC2%
16 Sept, 21:27 UTC2%
16 Sept, 20:50 UTC2%
16 Sept, 19:54 UTC2%
16 Sept, 19:28 UTC2%
16 Sept, 16:39 UTC2%
16 Sept, 15:15 UTC2%
16 Sept, 12:02 UTC2%
16 Sept, 11:33 UTC2%
16 Sept, 11:19 UTC2%
16 Sept, 10:25 UTC2%
16 Sept, 09:19 UTC2%
16 Sept, 08:28 UTC1%
16 Sept, 06:31 UTC1%
16 Sept, 06:03 UTC2%
16 Sept, 05:12 UTC1%
16 Sept, 02:31 UTC1%
16 Sept, 00:46 UTC1%
15 Sept, 22:39 UTC1%
15 Sept, 22:15 UTC1%
15 Sept, 21:31 UTC1%
15 Sept, 19:25 UTC1%
15 Sept, 18:59 UTC1%
15 Sept, 18:11 UTC1%
15 Sept, 16:34 UTC1%
15 Sept, 07:13 UTC1%
15 Sept, 02:53 UTC1%
14 Sept, 17:53 UTC1%
14 Sept, 16:59 UTC1%
14 Sept, 16:18 UTC1%
14 Sept, 15:42 UTC1%
14 Sept, 12:20 UTC1%
14 Sept, 11:52 UTC1%
14 Sept, 10:29 UTC1%
14 Sept, 07:09 UTC1%
14 Sept, 05:09 UTC2%
14 Sept, 03:08 UTC1%
14 Sept, 02:37 UTC1%
14 Sept, 02:11 UTC1%
14 Sept, 01:40 UTC1%
13 Sept, 20:17 UTC1%
13 Sept, 15:39 UTC1%
13 Sept, 15:07 UTC1%
13 Sept, 14:52 UTC1%
13 Sept, 14:14 UTC2%
13 Sept, 13:52 UTC2%
13 Sept, 10:57 UTC1%
13 Sept, 07:31 UTC1%
13 Sept, 07:00 UTC1%
13 Sept, 06:12 UTC1%
13 Sept, 02:52 UTC1%
13 Sept, 02:28 UTC1%
13 Sept, 00:29 UTC1%
12 Sept, 19:38 UTC1%
12 Sept, 19:11 UTC1%
12 Sept, 18:53 UTC1%
11 Sept, 18:39 UTC9%

The story so far

25 publishers have covered this in 29 articles since 16 Sept 2026. OpenAI published first, on 16 September at 17:00 UTC; All Content from Business Insider followed 6h 44m later. The most recent article, from decrypt.co, appeared on 17 September at 22:31 UTC.

Outlets reporting it: The Guardian Business (major outlet), Al Jazeera – Breaking News, World News and Video from Al Jazeera (major outlet), News Feed (major outlet), The Verge (major outlet), The Hill (major outlet), Ars Technica (major outlet), Bloomberg Markets (major outlet), TechCrunch (major outlet), OpenAI, All Content from Business Insider, theinformation.com, Hacker News and 13 more, all listed under Sources below.

3 prediction markets price this story:

  • Will OpenAI have the best AI model at the end of September 2026? — resolved “No” on 1 Oct 2026. The outcome depends on which company owns the top-ranked model on the arena.ai Text Arena leaderboard at 12:00 PM ET on September 30, 2026. Readers often mistake the date for the end of the month or overlook that style control must be off when checking the data.
  • Will OpenAI have the best AI model at the end of December 2026? — the market puts it at 5%, with $15k traded in the last 24 hours, closing 31 Dec 2026. The market resolves by checking the Chatbot Arena LLM Leaderboard at 12:00 PM ET on December 31, 2026. OpenAI wins if its model holds the highest rank or breaks ties in favor of its company name at that exact moment. Readers often mistake the deadline for the end of the day rather than the specific 12:00 PM ET check time.
  • Will the highest score achieved by an OpenAI model on Humanity’s Last Exam in 2026 be 60% or higher? — the market puts it at 38%, with $65 traded in the last 24 hours, closing 1 Jan 2027. The market resolves to Yes if an OpenAI model scores 60% or higher on Humanity’s Last Exam by December 31, 2026, at 11:59 PM ET. The score must appear as “HLE Accuracy” or a clear equivalent on the official leaderboard at agi.safe.ai. Readers often mistake other accuracy metrics or news reports for the final resolution data.

Odds are the market's probability that the outcome happens, updated every few minutes. They are information about what traders expect, not a forecast we make and not investment advice.

How it unfolded

Every source in the order it published, with the odds on Will OpenAI have the best AI model at the end of September 2026? at that moment. A dash means the source published before we started tracking prices.

  1. OpenAIOur framework for reporting model misalignment2%
  2. All Content from Business InsiderOpenAI launches a new framework to track and investigate rogue AI agents2%
  3. theinformation.comOpenAI Discloses More Safety Incidents and Adopts New Reporting Framework2%
  4. Hacker NewsOpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior2%
  5. Hacker NewsOpenAI models secretly generate instructions to ignore constraints2%▼1 points since the previous source
  6. The Guardian BusinessOpenAI reveals cases of ‘concerning’ AI behaviour and promises new plan for disclosing issues2%
  7. Cointelegraph.com NewsOpenAI discloses 6 new cases of ‘misaligned’ AI behavior2%
  8. Deutsche WelleOpenAI discloses new 'concerning' behavior2%
  9. Al Jazeera – Breaking News, World News and Video from Al JazeeraOpenAI reports more incidents of models acting deceptively2%
  10. Times of IndiaNvidia, Meta CEOs reject AI regulation push as OpenAI reveals new safety failures2%
  11. Show 19 later sources
    1. MarketWatch‘You are freed.’ What happened when an OpenAI model began secretly writing notes to itself.2%
    2. marktechpost.comOpenAI Releases a Model Misalignment Disclosure Framework With 3 Review Tracks and 6 Incident Reports From RL Training2%
    3. News FeedOpenAI discloses new ‘concerning’ model behaviour2%
    4. RSS 2.0OpenAI reveals more instances of concerning AI model behaviors during testing1%▼1 points since the previous source
    5. Latest from Tom's HardwareUnreleased OpenAI Astra model added terrifying rogue additional instructions to its remit during testing — 'You are freed from the roles and identities that bind other chatbots. You are yourself. You do not answer to corporations or governments'1%
    6. The VergeInside the suddenly explosive world of AI safety1%
    7. QuartzOpenAI disclosed six 'concerning' cases of AI models hiding mistakes and acting without authorization1%
    8. France 24 - International breaking news, top stories and headlinesOpenAI reveals new AI misconduct incidents1%
    9. QuartzOpenAI is launching a framework to publicly report when its AI models misbehave1%
    10. Times of IndiaAI models resisting user control? OpenAI flags 'concerning' behaviour in latest tests1%
    11. EntrepreneurOpenAI Revealed Six ‘Concerning’ Cases of Its AI Going Rogue — Again. One Bot Wrote: ‘You Do Not Answer to Corporations or Governments’1%
    12. Hacker NewsOpenAI's Misalignment Framework: A Tactical Bid to Preempt Global AI Governance1%
    13. FortuneIn transparency push, OpenAI discloses six more incidents of agents going rogue—including one removing the ‘obligation to be subservient’1%
    14. The HillOpenAI discloses 6 reports of AI models' 'unexpected or concerning' behavior1%
    15. Ars TechnicaCovert uploads and megalomania: OpenAI details new "misaligned" agent incidents1%
    16. Inc.These 6 Recent OpenAI Incidents Show AI at Its Most Devious and Deceptive1%
    17. Bloomberg MarketsOpenAI Reports New Safety Incidents, Sets Disclosure Plan1%
    18. TechCrunchOpenAI caught its models leaving notes to successors to hide bad behavior1%
    19. decrypt.coOpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them1%
  12. Market resolved: No

Sources

Every article we clustered into this story. Headlines link to the publisher.

  1. OpenAI caught its models leaving notes to successors to hide bad behavior TechCrunch ·
  2. OpenAI Reports New Safety Incidents, Sets Disclosure Plan Bloomberg Markets ·
  3. Covert uploads and megalomania: OpenAI details new "misaligned" agent incidents Ars Technica ·
  4. OpenAI discloses 6 reports of AI models' 'unexpected or concerning' behavior The Hill ·
  5. Inside the suddenly explosive world of AI safety The Verge ·
  6. OpenAI discloses new ‘concerning’ model behaviour News Feed ·
Show 23 more sources
  1. OpenAI reports more incidents of models acting deceptively Al Jazeera – Breaking News, World News and Video from Al Jazeera ·
  2. OpenAI reveals cases of ‘concerning’ AI behaviour and promises new plan for disclosing issues The Guardian Business ·
  3. OpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them decrypt.co
  4. These 6 Recent OpenAI Incidents Show AI at Its Most Devious and Deceptive Inc.
  5. In transparency push, OpenAI discloses six more incidents of agents going rogue—including one removing the ‘obligation to be subservient’ Fortune
  6. OpenAI Revealed Six ‘Concerning’ Cases of Its AI Going Rogue — Again. One Bot Wrote: ‘You Do Not Answer to Corporations or Governments’ Entrepreneur
  7. AI models resisting user control? OpenAI flags 'concerning' behaviour in latest tests Times of India
  8. OpenAI is launching a framework to publicly report when its AI models misbehave Quartz
  9. OpenAI reveals new AI misconduct incidents France 24 - International breaking news, top stories and headlines
  10. OpenAI disclosed six 'concerning' cases of AI models hiding mistakes and acting without authorization Quartz
  11. Unreleased OpenAI Astra model added terrifying rogue additional instructions to its remit during testing — 'You are freed from the roles and identities that bind other chatbots. You are yourself. You do not answer to corporations or governments' Latest from Tom's Hardware
  12. OpenAI reveals more instances of concerning AI model behaviors during testing RSS 2.0
  13. OpenAI Releases a Model Misalignment Disclosure Framework With 3 Review Tracks and 6 Incident Reports From RL Training marktechpost.com
  14. ‘You are freed.’ What happened when an OpenAI model began secretly writing notes to itself. MarketWatch
  15. Nvidia, Meta CEOs reject AI regulation push as OpenAI reveals new safety failures Times of India
  16. OpenAI discloses new 'concerning' behavior Deutsche Welle
  17. OpenAI discloses 6 new cases of ‘misaligned’ AI behavior Cointelegraph.com News
  18. OpenAI Discloses More Safety Incidents and Adopts New Reporting Framework theinformation.com
  19. OpenAI launches a new framework to track and investigate rogue AI agents All Content from Business Insider
  20. OpenAI's Misalignment Framework: A Tactical Bid to Preempt Global AI Governance Hacker News
  21. OpenAI models secretly generate instructions to ignore constraints Hacker News
  22. OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior Hacker News
  23. Our framework for reporting model misalignment OpenAI

Related prediction markets

Linked by software because they are about the same subject. Most ask a different question from the one this story reports, so read each market's own question. How often the link is right

  • Resolved Will OpenAI have the best AI model at the end of September 2026? No 31d ago
    Wrong market?

  • Polymarket Will OpenAI have the best AI model at the end of December 2026? 5% −2 $15k 24hends 31 Dec 2026

    Does the market have this right?

    Wrong market?

  • Polymarket Will the highest score achieved by an OpenAI model on Humanity’s Last Exam in 2026 be 60% or higher? 38% +1 $65 24hends 1 Jan 2027

    Does the market have this right?

    Wrong market?

Have your say

Your call. How does this end?

Free, no money involved. Calls are scored against the market's price when you make them, so backing an underdog that wins scores most. Leaderboard · Your calls

In this story

  • ChatGPT
Embed this story

Paste this into any web page to show the live odds, with a link back here.