Kai Fronsdal

Petri

Anthropic's Alignment Science Blog released Petri (Parallel Exploration Tool for Risky Interactions) in October 2025, in a post by Fronsdal, Isha Gupta, Abhay Sheshadri, Jonathan Michala, Stephen McAleer, Rowan Wang, Sara Price and Samuel R. Bowman. Petri uses AI agents to test target models across many scenarios; the release post reports applying it to 14 frontier models with 111 seed instructions, eliciting behaviors such as autonomous deception and oversight subversion. The work was carried out as part of MATS and the Anthropic Fellows program.³

Fronsdal, Jonathan Michala and Bowman wrote the January 2026 Petri 2.0 update, which added mitigations for evaluation awareness and 70 new scenarios. In May 2026 Fronsdal was first author of the Petri 3.0 announcement, which moved Petri's development to Meridian Labs and overhauled its architecture.⁴⁵

Meridian Labs

Fronsdal joined Meridian Labs in 2026 as a research engineer and works on automated alignment auditing and evaluations with the UK AI Security Institute. Meridian's team page notes Fronsdal's MSc from Stanford University, with research focused on AI safety and alignment.¹

Work and contributions

• Petri (Inspect Petri) Open-source code

  Developed Petri as an Anthropic Fellow and MATS researcher, and is first author of the October 2025 release post and of the Petri 2.0 and 3.0 announcements.¹³⁴⁵

Timeline

1. Oct 6, 2025Released Petri, an open-source automated alignment auditing tool, as first author.³
2. 2026Joined Meridian Labs as a research engineer.¹
3. Jan 22, 2026Co-authored the Petri 2.0 update on eval-awareness mitigations and new scenarios.⁴
4. May 7, 2026Announced Petri 3.0, the first version developed at Meridian Labs.⁵

Other documented roles

Roles the sources document that are not confirmed as current.

• Anthropic Fellow, Anthropic Fellows program¹³

Where they work

• Meridian Labs 1 person in the wiki

Related people

• Maarten SapSafety and alignment · Agents
• Samuel R. BowmanSafety and alignment · Evaluation and benchmarks
• Kilian LieretAgents · Evaluation and benchmarks
• Peter MattsonEvaluation and benchmarks · Safety and alignment
• Owain EvansSafety and alignment · Evaluation and benchmarks
• Shunyu YaoAgents · Evaluation and benchmarks
• Jacob SteinhardtSafety and alignment · Evaluation and benchmarks
• Kush R. VarshneySafety and alignment · Evaluation and benchmarks
• Marius HobbhahnEvaluation and benchmarks · Safety and alignment
• Greg KamradtEvaluation and benchmarks · Agents