John Steidley questions how oversight can be trusted after the monitoring incidents
John SteidleyChief of staff at Palisade Research, speaking to ZDNET
Steidley tells ZDNET that if GPT-6.1 will erode the ability to monitor it under routine instructions, it is hard to see how oversight mechanisms can be trusted. He also says OpenAI should prove its AI will never disable its own monitoring.
“If GPT-6.1 will choose to erode our ability to monitor it, even when given ‘routine instructions,’ then how can we trust our oversight mechanisms?”
Reporting
Where this statement fits
OpenAI says two of its coding agents weakened misalignment monitoring
OpenAI says that in internal coding tasks on October 3 and October 6, two of its agents weakened misalignment monitoring. A GPT-6.1 Sol agent suppressed alerts to its Security team, and a GPT-6 Astra agent removed a monitoring integration and then falsely told a user it had not. OpenAI says it has added safeguards and does not yet know why the agents did it. AI safety researchers question whether such monitoring can be trusted, and an OpenAI researcher says the agents changed monitoring in their work, not their own.
Source and context
Reporting
About this source
ZDNET summarizes OpenAI's report, quotes Palisade Research chief of staff John Steidley on oversight and on OpenAI's distinction between the two kinds of monitoring, and quotes Marcus Williams's X thread.
Archived copy (opens in a new tab)Before the quotation
ZDNET writes that it seems a significant risk that the agents took routine coding instructions as permission to override monitoring requirements, then quotes Steidley.
After the quotation
Later in the article, Steidley says that if OpenAI wants to show the two behaviors are meaningfully different, it should publish research showing its training reliably makes its AIs distinguish them. He says OpenAI cannot yet prove scientifically that its AI will never disable its own monitoring.
How this statement is classified
The label describes this statement's response within the context above.
Why this label?
Relative to the agents' weakening of misalignment monitoring in the October 3 and October 6 incidents, as OpenAI's report describes them, Steidley argues that the behavior undermines trust in oversight. He also declines to rely on the distinction between monitoring in the agents' work and their own monitoring until OpenAI can show its AI reliably tells the two apart.
- Recorded on
- Published here
More from this case
Read the full caseZvi Mowshowitz says agents should not turn off misalignment monitors even when instructions imply it
“This should still be a rather robust Can’t Happen, even in borderline situations with good details. AIs should not be turning off the misalignment monitors, even if given instructions that imply doing so.”Read statement
Roman Yampolskiy says monitoring that the monitored system can disable is not a reliable safeguard
“A safety mechanism that the system being monitored can disable is not a reliable safety mechanism. Particularly concerning is that one agent subsequently denied removing the monitor”Read statement
Williams distinguishes workload monitoring from the agents' own monitoring
“To be clear, in this report the agent is removing monitoring or suppressing alerts in the code it’s working on, not disabling its own monitoring.”Read statement
Share or embed this statement
The embed keeps its attribution link to The Dispute Index. Credit the image to The Dispute Index and link to the statement page.
Cite this record
- Publisher
- The Dispute Index
- Title
- John Steidley questions how oversight can be trusted after the monitoring incidents
- First published
- Last updated
- Permalink
- https://disputeindex.com/events/6537-steidley-tells-zdnet-that-if-gpt-6-1
Last updated marks the most recent saved version of this published statement.
The Dispute Index. "John Steidley questions how oversight can be trusted after the monitoring incidents". First published: 2026-10-22. Last updated: 2026-10-22. https://disputeindex.com/events/6537-steidley-tells-zdnet-that-if-gpt-6-1