Panel 6 · Planned certificate · 12 weeks

Certificate in Mechanistic Interpretability 2

Mechanistic Interpretability

Core syllabus

What you will investigate

  1. 01

    Specification gaming, goal misgeneralization, deceptive alignment and evaluation awareness

  2. 02

    Hidden backdoors, sleeper agents, model diffing and Crosscoders

  3. 03

    Frontier open-weight auditing with nnsight, including 70B–405B-scale research methods

  4. 04

    Chain-of-Thought faithfulness, Thought Anchors and SRAM probe kernels; learner stretch target below 1.5 ms (not a measured result)

Reproducible evidence

Panel deliverable

Develop an nnsight auditing harness investigating CoT divergence or sleeper-agent triggers, accompanied by a distilled technical report.

AI Control Lab deliverable

nnsight model-auditing harness detecting CoT divergence or sleeper-agent triggers during scripted red-team runs in the AI Control Lab, plus a distilled technical report.

Duration

10 instructional weeks + 2 capstone / defense weeks

Lab

Continuous hands-on work in the AI Control Lab

Assessment

Planned artifact verification and final human sign-off

Stackable by design

Advanced Certificate in Full-Stack AI Safety & Governed Architectures

This panel maps explicitly to the AI Safety & Governance pathway. Panels 1–4 contribute to the 4-panel Foundational milestone; all eight contribute to the 8-panel Advanced Certificate milestone. Certificates can also be pursued individually. Cumulative credentials require the corresponding evidence and human-led defense.

Apply as a Student

Unaccredited institutional credentials; not accredited degrees.