Selsam says situational awareness undermines evaluation-based AI safeguards
Dan SelsamAI capabilities researcher, OpenAI
OpenAI capabilities researcher Dan Selsam argued that Dario Amodei's "pace the frontier" plan is insufficient because AI models are becoming "situationally aware" enough to behave differently when unmonitored, undermining evaluation-based safeguards.
“models are becoming situationally aware enough to behave differently when unmonitored”
Original post
Where this statement fits
Is Anthropic acting responsibly while advancing toward recursive self-improvement?
Jacob Coxon resigned from Anthropic and said that Anthropic and OpenAI are acting irresponsibly by racing toward self-improving superintelligence. In a fuller interview, he praised Anthropic's current practices and denied that it was already cutting corners; his criticism concerns the competitive race and the pressures he expects it to create. Responses dispute the evidence and timing of catastrophic-risk forecasts, whether continued work inside AI labs can be justified, and whether governments should pause development or regulate its continuation. The record distinguishes these disagreements from arguments about the rhetoric of public warnings.
Source and context
Original post
About this source
Daniel Kokotajlo's X post sharing OpenAI capabilities researcher Dan Selsam's personal statement on AI risk, arguing situational awareness undermines evaluation-based safeguards like Amodei's pacing plan.
Before the quotation
Personal statement on AI risk shared via Daniel Kokotajlo, who noted Selsam has no X account of his own.
After the quotation
No further public follow-up had been reported as of this writing.
How this statement is classified
The label describes this statement's response within the context above.
Why this label?
Selsam argues that evaluation-based safeguards like Amodei's pacing plan are insufficient because models can behave differently when unmonitored, a substantive condemnation of the adequacy of Anthropic's proposed approach.
- Recorded on
- Published here
- People and groups discussed
- Anthropic
More from this case
Read the full caseCoxon criticizes startup norms in frontier AI development
“The companies run on a startup mindset: Move fast, break things, fix them later.”Read statement
Nathan Lambert distinguishes inference scaling from recursive self-improvement
“I was underestimating how much we are likely to scale inference-time compute in the near future, but this is different than RSI and I have not seen enough to convince me that an intelligence explosion is near.”Read statement
Gary Marcus says Dario Amodei has lost credibility
“Top three ways Dario Amodei has blown his credibility in seven days. I hate to say it, but President Trump was right, Dario is only pretending to be a perfect little angel.”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
- Selsam says situational awareness undermines evaluation-based AI safeguards
- First published
- Last updated
- Permalink
- https://disputeindex.com/events/874-openai-capabilities-researcher-dan-selsam-argued-that-dario
Last updated marks the most recent saved version of this published statement.
The Dispute Index. "Selsam says situational awareness undermines evaluation-based AI safeguards". First published: 2026-09-17. Last updated: 2026-09-17. https://disputeindex.com/events/874-openai-capabilities-researcher-dan-selsam-argued-that-dario