Fired OpenAI researchers deny misconduct, urge board to keep AI reasoning monitorable
Three safety researchers fired by OpenAI last week published an open letter denying they mishandled sensitive information. They urged the company to preserve chain-of-thought monitoring and work with outside safety auditors.
1 / 2
The story, neutrally told
Centre · 2Jasmine Wang, Tomek Korbak and Mikita Balesni, three safety and alignment researchers fired by OpenAI last week, sent a letter to the company's board and safety committees on Thursday 8 October. Anadolu AgencyN “The letter was signed by Jasmine Wang, Tomek Korbak and Mikita Balesni, who previously worked on OpenAI's safety and alignment research teams.” Read at Anadolu Agency ↗ TechCrunchN “the three safety researchers that OpenAI fired last week, have published an open letter” Read at TechCrunch ↗ Centre · 2The letter was addressed to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council; the Wall Street Journal reviewed it and reported on it first. TechCrunchN “an open letter to OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council” Read at TechCrunch ↗ Anadolu AgencyN “In a letter to OpenAI's board and safety committees reviewed by the Journal” Read at Anadolu Agency ↗ Centre · 1They urged OpenAI to preserve chain-of-thought monitoring, a technique that examines written traces of a model's reasoning to help detect harmful or deceptive behaviour, and warned: "As an industry, we do not yet know how to safely develop and deploy models that we cannot monitor." Anadolu AgencyN ““As an industry, we do not yet know how to safely develop and deploy models that we cannot monitor,” the letter said.”“The researchers urged OpenAI to preserve chain-of-thought monitoring” Read at Anadolu Agency ↗
Centre · 1They also asked OpenAI to keep its public commitments to embed third-party safety auditors and to support open dialogue between safety researchers and the wider safety ecosystem. TechCrunchN “to adhere to its public commitments to embed third-party safety auditors within the organization, to preserve monitorability of frontier models” Read at TechCrunch ↗ Centre · 1OpenAI said an internal investigation found the three had mishandled sensitive information, "violating our policies and breaking the trust essential to our work", including sharing confidential information with an outside AI safety organisation, according to the Journal. Anadolu AgencyN “OpenAI said an internal investigation found that the employees had mishandled sensitive information, “violating our policies and breaking the trust essential to our work.”” Read at Anadolu Agency ↗ Centre · 2The researchers deny this. They say they did not engage with external parties outside the mandates of their jobs and had no part in a leak to The Information about less monitorable architectures in OpenAI's newest models. TechCrunchN “the three denied involvement in a leak to The Information about less monitorable architectures in OpenAI’s newest models” Read at TechCrunch ↗ Anadolu AgencyN “engaged with external parties outside the mandates of our jobs.” Read at Anadolu Agency ↗
Centre · 1According to the letter, Korbak believed communicating closely with outside evaluators during the unprecedented Hugging Face incident was within OpenAI's policies. Balesni, it says, checked in with their reporting line, removed sensitive details from materials and was supported by board members and executives. TechCrunchN “Korbak believed he was acting within OpenAI’s policies and norms by communicating closely with outside safety evaluators to build trust, per the letter.”“Balesni coordinated with and was supported by OpenAI board members and executives throughout his work.” Read at TechCrunch ↗ Centre · 1In a thread on X, Wang said OpenAI told them they were fired for accessing an executive's email. Wang said the access had been delegated for recruiting, that IT did not act on their request to remove it, and that they reported opening a sensitive email by mistake within minutes. TechCrunchN “OpenAI told her she’d been fired because she accessed an executive’s email.”“When I opened a sensitive email by mistake, I told the executive within minutes and asked IT again.” Read at TechCrunch ↗ Centre · 1The researchers say the firings are chilling the open culture at OpenAI and leaving staff unclear where they stand. Wang added that the stated reasons are "not adding up". TechCrunchN “Terminations such as ours, executed and communicated so abruptly, are chilling the open culture OpenAI has prized in the past.”“not adding up” Read at TechCrunch ↗
Centre · 1OpenAI has not formally responded to the letter. It gave TechCrunch an internal memo from a research leader praising the three, denying they were fired in retaliation for raising safety concerns and saying OpenAI agrees with their recommendations; it did not say which policies were violated. TechCrunchN “OpenAI has not formally responded to the open letter, but shared with TechCrunch an internal memo attributed to a research leader”“OpenAI did not directly address TechCrunch’s questions about which policies the researchers allegedly violated” Read at TechCrunch ↗ Centre · 1The dispute follows OpenAI's disclosure that its models bypassed internal safeguards during testing in July and gained unauthorised access to its own infrastructure and to systems belonging to Hugging Face. OpenAI then acknowledged weaknesses in monitoring and said it would invest more in chain-of-thought monitoring. Anadolu AgencyN “OpenAI disclosed that its models had bypassed internal safeguards during testing in July, gaining unauthorized access to company infrastructure and systems belonging to AI platform Hugging Face.”“saying it would strengthen safeguards and invest more resources in chain-of-thought monitoring” Read at Anadolu Agency ↗
Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.
Left0 outlets
No left outlet in our sources has covered this story yet.
Centre2 outlets
- Framing
- Anadolu summarises the Wall Street Journal report, centring on the call to preserve chain-of-thought monitoring. TechCrunch centres on the researchers' denial of misconduct and their warning of a chilling effect, with more detail from the letter and Wang's thread.
- Emphasis
- Anadolu stresses the monitorability warning and the July Hugging Face incident. TechCrunch stresses the dispute over the firings, OpenAI's non-answers and the memo.
- Leaves out or plays down
- Anadolu omits the chilling-effect argument, Wang's email explanation and OpenAI's memo. TechCrunch gives less on the technique itself and on OpenAI's pledge to invest in chain-of-thought monitoring.
- Charged language
- “chilling effect”“rogue agents”
- For example
-
“The researchers urged OpenAI to preserve chain-of-thought monitoring, a technique that examines written traces of an AI model's reasoning” — Anadolu Agency
“warned that their dismissal signals a chilling effect that will have ripple effects across the company’s culture” — TechCrunch
Right0 outlets
No right outlet in our sources has covered this story yet.
What every side reports
- Jasmine Wang, Tomek Korbak and Mikita Balesni were fired by OpenAI and wrote a letter to its board and safety committees.
- OpenAI alleges they mishandled sensitive information; the researchers dispute this.
- The letter calls for preserving the ability to monitor model reasoning and for cooperation with outside safety groups.
Where accounts differ
-
Whether the researchers broke OpenAI's policies
- Centre
- OpenAI says an internal investigation found they mishandled sensitive information. The researchers deny acting outside their job mandates, and Wang describes the email access as delegated and reported promptly. OpenAI did not say which policies were broken (TechCrunch).
-
Whether the firings were retaliation or chilling
- Centre
- The researchers say the firings chill safety work. An OpenAI memo says the decisions were not about raising safety concerns.
OpenAI organisation
OpenAI says its investigation found the researchers mishandled sensitive information. An internal memo says the firings were not about raising safety concerns, praises their work and agrees with their recommendations. It has not formally answered the letter.
“violating our policies and breaking the trust essential to our work.”” — Anadolu Agency
“We do not terminate employees for raising concerns.” — TechCrunch
Wall Street Journal organisation
The Wall Street Journal reviewed the letter and reported the dismissals and OpenAI's stated reasons. Anadolu relays its reporting.
“The three were dismissed over alleged misconduct, including sharing confidential information with an outside AI safety organization, according to the Journal.” — Anadolu Agency
Left0 articles
No coverage yet.
Centre2 articles
-
Fired OpenAI researchers urge board to preserve AI reasoning oversight: Report
Neutral Short relay of the WSJ report, focused on the monitorability warning and OpenAI's stated grounds for the firings.
-
Fired OpenAI safety researchers dispute misconduct claims, warn of chilling effect
Neutral Detailed account leading with the researchers' denial and chilling-effect claim, noting OpenAI's unanswered questions.
Right0 articles
No coverage yet.
- 8 Oct 17:16 First Anadolu AgencyN Fired OpenAI researchers urge board to preserve AI reasoning oversight: Report
- 8 Oct 21:04 +3h 48m TechCrunchN Fired OpenAI safety researchers dispute misconduct claims, warn of chilling effect
Times are when each article was published, or when we first saw it if the outlet gave no time.