Head of Safety

Cognition · San Francisco · Posted Oct 9, 2026


Transparency

Dead

Online for 6 weeks and never reposted. Jobs this old or reposted this often rarely lead to a hire. It may be filled already.

First seen
Oct 10, 2026
Offered for
42 days

Overview

Category
AI & machine learning
Employment type
Permanent
Workplace type
On-site
Country
United States
Seniority
Head of
Schedule
Full time

Key skills

Safety strategy development
Define and own the overall safety approach and outcomes for the autonomous agent.
Evaluation and red‑team design
Create and integrate evals and red‑team tests for agentic risks into each release.
Agentic risk mitigation
Identify and address failure modes like tool misuse, prompt injection, and reward hacking.
Deployment policy definition
Set permissions, oversight, and usage guidelines for the agent in enterprise contexts.
Safety team building
Hire, lead, and grow researchers and engineers focused on AI safety.
Python and ML stack fluency
Read, modify, and run code in the training and inference pipelines for safety work.

Key phrases

  • agentic risk
  • unsafe actions
  • prompt injection
  • data exfiltration
  • sandbox escape
  • reward hacking
  • misuse
  • deployment policy
  • enterprise customers
  • published or widely used safety research

From the posting

We are an applied AI lab building end-to-end software agents.

We're the makers of Devin, the first AI software engineer.

Our team is extremely talent-dense. Among our founding team, we have world-class competitive programmers, former founders, and leaders from companies at the cutting edge of AI including Scale AI, Palantir, Cursor, Waymo, Tesla, Lunchclub, Modal, Google DeepMind, and Nuro.

Building Devin is just the first step—our hardest challenges still lie ahead. If you’re excited to solve some of the world’s biggest problems and build AI that can reason on real-world tasks, apply to join us.

Role Mission

Devin is one of the most capable autonomous agents in production. It writes and runs code, uses tools, takes actions in real customer systems, and operates for hours without a human in the loop. That makes it one of the most important places in the industry to get safety right, and one of the few where safety work ships to millions of developers rather than staying in a paper. You will build Cognition's safety function from the ground up: the research agenda, the evaluation and red-teaming program, the deployment policies, and the team. You will work directly with the researchers training our models and the engineers building the agent harness. This is a hands-on role for someone who has done serious safety or alignment work at a frontier lab and wants to apply it where agents actually operate.

What You'll Accomplish

  • Own safety end to end: Set the safety strategy for Devin and the models behind it, and own the outcomes.

  • Build the evaluation program: Design evals and red-teaming for agentic risk: unsafe actions, prompt injection, data exfiltration, sandbox escape, reward hacking, and misuse. Make them part of every model and product release.

  • Shape training and the agent harness: Partner with pre-training, post-training, and agent teams so safety is built into how models are trained and how Devin plans, acts, and asks for help.

  • Set deployment policy: Define what Devin is and is not allowed to do, how permissions and oversight work, and how we handle the gray areas, in a way that holds up with enterprise customers.

  • Build the team and the external voice: Hire and lead safety researchers and engineers. Represent Cognition's safety work with customers, policymakers, and the broader research community.

Exceptional Candidates Have Demonstrated

  • Frontier lab safety or alignment experience: Hands-on work on safety, alignment, evaluations, or red-teaming at a frontier AI lab. You have shipped safety work that affected real models or products.

  • Agentic systems depth: You understand how agents fail in practice: tool misuse, specification gaming, long-horizon drift, adversarial inputs. You have ideas about how to measure and mitigate it.

  • Research credibility: A track record of published or widely used safety research, evals, or methods. An advanced degree in Computer Science, Machine Learning, or a related field is a plus.

  • Technical fluency: Comfortable in Python and in the training and inference stack. You can read the code, run the evals, and argue with researchers on the details.

  • Builder, not reviewer: You have started or scaled a safety function and prefer shipping mitigations to writing memos about them.

  • Judgment under uncertainty: You can make clear calls on deployment risk with incomplete information and explain them to engineers, executives, and customers.

Our Commitment

Equal Opportunity. Cognition is an equal opportunity employer. We do not discriminate on the basis of any protected characteristic under applicable law. We are committed to providing reasonable accommodations for candidates with disabilities throughout the hiring process - please let us know if you need any. Cognition also considers applicants with criminal histories consistent with the SF Fair Chance Ordinance and similar local laws.

Data Privacy and Arbitration: By submitting your application, you agree to our Global Applicant and Job Candidate Privacy Notice. For US applicants, by submitting your application, you agree to the Applicant Arbitration Agreement.

More jobs at Cognition

Recruiter - UK

Cognition · 2 days ago | London | On-site · Senior · Full time

Create CV

Recruiting Sourcer - UK

Cognition · 2 days ago | London | On-site · Mid level · Full time

Create CV

Workplace & Facilities Lead (West Coast)

Cognition · Last week | San Francisco | On-site · Lead · Full time

Create CV

Search & AI Discoverability Lead

Cognition · Last week | New York City | On-site · Lead · Full time

Create CV

Lifecycle Marketing Lead

Cognition · Last week | San Francisco | On-site · Lead · Full time

Create CV
See all 86 jobs at Cognition