Josh Clymer

Commentary by Josh Clymer

From the commentary rail — every link leaves the site for the original piece.

• 15 November 2025 · Redwood ResearchWill AI systems drift into misalignment?A reason alignment could be hard
• 14 July 2025 · Redwood ResearchRecent Redwood Research project proposalsEmpirical AI security/safety projects across a variety of areas
• 25 April 2025 · Redwood ResearchClarifying AI R&D threat models(There are a few)
• 19 February 2025 · Redwood ResearchHow might we safely pass the buck to AI?Developing AI employees that are safer than human ones
• 30 January 2025 · Redwood ResearchTakeaways from sketching a control safety caseInsights from a long technical paper compressed into a fun little commentary
• 29 January 2025 · Redwood ResearchPlanning for Extreme AI RisksAre we ready for this?
• 22 January 2025 · Redwood ResearchWhen does capability elicitation bound risk?The assumptions behind and limitations of capability elicitation have been discussed in multiple places (e.g.
• 13 January 2025 · Redwood ResearchExtending control evaluations to non-scheming threatsBuck Shlegeris and Ryan Greenblatt originally motivated control evaluations as a way to mitigate risks from ‘scheming’ AI models: models that consistently pursue power-seeking goals in a covert way; however, many adversarial model psychologies are not well described by the standard notion of…