---
title: "Research Scientist, Alignment & Mechanistic Interpretability"
slug: "research-scientist-alignment-mechanistic"
id: "3316318b-1e61-4cc4-9761-691dcb9477f9"
department: "AI Research"
department_code: "RES"
location: "San Francisco, CA"
workplace_type: "Hybrid"
employment_type: "Full-time"
compensation: "$280k – $380k (annual package)"
is_remote: false
is_featured: false
headcount: 1
canonical_url: "https://jobs.talent247.co/jobs/research-scientist-alignment-mechanistic"
markdown_url: "https://jobs.talent247.co/jobs/research-scientist-alignment-mechanistic.md"
apply_endpoint: "https://jobs.talent247.co/api/public/jobs/research-scientist-alignment-mechanistic/apply"
schema_endpoint: "https://jobs.talent247.co/api/public/jobs/research-scientist-alignment-mechanistic/form-schema"
webmcp_tool: "apply_for_role"
updated_at: "2026-11-15T17:37:56.522Z"
content_signal: "search=yes, ai-input=yes, ai-train=no"
---

# Research Scientist, Alignment & Mechanistic Interpretability

## Position Telemetry

- **Reference:** `RESEARCH-SCIENTIST`
- **Practice / Sector:** AI Research (`RES`)
- **Location:** San Francisco, CA
- **Workplace Type:** Hybrid
- **Employment Type:** Full-time
- **Compensation:** $280k – $380k (annual package)
- **Canonical HTML URL:** https://jobs.talent247.co/jobs/research-scientist-alignment-mechanistic

## Role Dossier & Specification

THE CLIENT & MANDATE
Our client is a mission-driven frontier AI research laboratory headquartered in San Francisco, renowned globally for pioneering constitutional alignment, mechanistic interpretability, and frontier coding and reasoning models. Their Interpretability & Alignment Science organization seeks to reverse-engineer the internal computational circuits of multi-hundred-billion-parameter transformers so safety guarantees rest on mechanistic understanding rather than black-box behavioral spot checks alone.

They have retained Talent247 to appoint a Research Scientist in Alignment & Mechanistic Interpretability in San Francisco, CA. Working alongside pioneers of dictionary learning and attribution graphs, the appointed scientist will scale interpretability decompositions to flagship production models and translate circuit-level discoveries into concrete training-time and inference-time alignment interventions.

KEY RESPONSIBILITIES
• Originate and scale dictionary-learning architectures (sparse autoencoders, transcoders, cross-layer attribution graphs) across 70B+ to frontier-scale multimodal and reasoning transformers.
• Trace internal computational circuits governing multi-step planning, chain-of-thought faithfulness, sycophancy, sandbagging, reward hacking, and refusal behavior.
• Design causal activation-patching, feature-steering, and representation-monitoring interventions that detect deceptive or misaligned latent states during online RL and real-world agent deployment.
• Partner with distributed systems engineers to build petabyte-activation caching and low-overhead auxiliary dictionary-training pipelines integrated directly into flagship training clusters.
• Construct rigorous quantitative benchmarks evaluating whether interpretability tools genuinely improve human/automated auditing of subtle sleeper-agent or specification-gaming model organisms.
• Publish foundational open research advancing the broader AI safety and mechanistic interpretability community.

TECHNICAL ENVIRONMENT & OFFERING
Stack: Python, PyTorch, JAX, Triton custom sparse-dictionary kernels, distributed activation-streaming pipelines, multi-thousand GPU/ASIC research clusters. Workplace: Hybrid in San Francisco, CA. Compensation: $280,000 – $380,000 base salary plus generous frontier-lab equity, comprehensive health & wellness benefits, and relocation/visa support.

## Core Requirements

ESSENTIAL QUALIFICATIONS
• PhD in Machine Learning, Physics, Neuroscience, Mathematics, Computer Science, or equivalent frontier research track record in interpretability, alignment, or deep learning science.
• Demonstrated originality through influential papers, technical reports, or open-source work in mechanistic interpretability, representation engineering, scalable oversight, or alignment stress-testing.
• Strong hands-on fluency in PyTorch or JAX—comfortable writing clean distributed experiments over large transformer internals (residual streams, attention heads, MLP activations).
• Deep scientific skepticism: habit of testing competing null hypotheses and designing falsifiable causal ablations rather than relying on cherry-picked qualitative examples.
• Collaborative communication style suited to cross-pollinating fundamental alignment science with production post-training teams.

PREFERRED / BONUS EXPERIENCE
• Direct experience training sparse autoencoders (SAEs), cross-layer transcoders, or circuit-attribution pipelines on large language models.
• Experience constructing alignment "model organisms" to stress-test auditing, unlearning, or RLHF robustness techniques.
• Familiarity with custom Triton/CUDA sparse matrix kernels for scaling million-feature dictionary learning.

## Application Schema & Direct Apply (`apply_for_role`)

Candidates and authorized AI agents on behalf of a candidate can submit an application via the declarative WebMCP form tool `apply_for_role` on https://jobs.talent247.co/jobs/research-scientist-alignment-mechanistic or via HTTP `POST https://jobs.talent247.co/api/public/jobs/research-scientist-alignment-mechanistic/apply` (`application/json` or `multipart/form-data`). Non-browser HTTP agents must include header `X-Talent247-Intake-Token` obtained from `GET https://jobs.talent247.co/api/public/intake-token` (or automatically returned in `GET https://jobs.talent247.co/api/public/jobs/research-scientist-alignment-mechanistic/form-schema`).

### Form Fields

- `full_name` (text, required): Full name
- `email` (email, required): Email
- `phone` (phone, optional): Phone
- `location` (text, optional): Location
- `linkedin_url` (url, optional): LinkedIn URL
- `cover_letter` (textarea, optional): Cover letter
- `consent` (checkbox, required): I consent to my data being processed for this application
- `cv_file` / `cv_base64` + `cv_filename`: Upload a CV in .pdf or .docx format (<= 10 MB) or provide an `https://` LinkedIn profile URL in `linkedin_url`.

### Alternative: Talent Pool Intake (`join_talent_pool`)

If this mandate is not an exact match, register once in the Talent247 Talent Pool at https://jobs.talent247.co/talent-pool (`https://jobs.talent247.co/talent-pool.md`) via WebMCP tool `join_talent_pool` or `POST https://jobs.talent247.co/api/public/talent-pool`.
