Build safer agents.
Protect what matters.
As we gain scale and momentum, we would be delighted to collaborate with talented people across the areas below. Share your interests and experience, and help us build a future where businesses can adopt AI agents safely.
Explore 15 areas Find where you can make an impact.Explore future opportunities.
FIND YOUR FOCUS
Where you can make an impact.
Explore the areas where we hope to collaborate as we grow. Registering your interest helps us connect when the right opportunity becomes available.
Our locations: San Francisco, New York, Singapore, Poland.
Research / 01AI Safety Researcher
Study how autonomous agents behave when they encounter conflicting goals, untrusted content and sensitive information.
Scope and responsibilities
- Design evaluations for agents that handle sensitive information, use external tools and encounter conflicting goals.
- Build reproducible experiments with explicit baselines, documented assumptions and meaningful measures of harmful behaviour.
- Translate findings into product safeguards and collaborate with engineers to test whether those safeguards work.
What we would look for
- Practical experience in empirical ML research, model evaluation or a closely related quantitative discipline.
- Strong Python and experiment design skills, including the ability to investigate noisy or inconsistent results.
- Clear technical writing and a willingness to distinguish demonstrated risks from speculation and unresolved questions.
The challenges
Agent behaviour varies across models, prompts and environments. You would help build evaluations that remain useful as capabilities change, while avoiding conclusions that extend beyond the evidence.
Register interestResearch / 02AI Security Researcher
Investigate how agent tools, MCP integrations and model behaviour create new security risks.
Scope and responsibilities
- Investigate attack surfaces across agent tools, MCP integrations, authentication flows and connected business systems.
- Develop controlled proofs of concept, reproduce findings and assess the limits of proposed mitigations.
- Work with product and engineering teams to turn research into prioritised fixes and repeatable security checks.
What we would look for
- Hands-on experience with application security research, adversarial testing or security assessments of AI systems.
- An ability to trace data and authorisation across several components rather than treating a model in isolation.
- Careful handling of sensitive information and clear documentation of impact, prerequisites and responsible disclosure decisions.
The challenges
The same integration can behave differently depending on its permissions, deployment and downstream services. The challenge is to identify exploitable conditions without confusing a theoretical weakness with a demonstrated security failure.
Register interestSecurity / 03Security Specialist
Help teams understand and reduce the risks in their AI agent deployments.
Scope and responsibilities
- Review agent configurations, identities, tool access and credential handling within an explicitly agreed assessment scope.
- Collect evidence, prioritise findings and propose practical changes that fit the customer’s working environment.
- Help develop consistent assessment checklists and verify whether remediation has addressed the original issue.
What we would look for
- A solid grounding in application security, identity and access management, and common cloud or enterprise integrations.
- Experience writing actionable findings for technical teams and explaining their business relevance without overstating impact.
- Sound judgement when evidence is incomplete, access is restricted or a proposed test could affect a live system.
The challenges
Every organisation has different ownership, access boundaries and operational constraints. You would need to uncover meaningful risks while keeping assessments focused, authorised and understandable to the people responsible for fixing them.
Register interestEngineering / 04DevSecOps Engineer
Build security into the systems and workflows used to develop and deliver aioscope.
Scope and responsibilities
- Build secure delivery workflows with dependency checks, infrastructure validation and appropriate controls around releases.
- Improve secrets management, environment isolation and the consistency of development, testing and production configuration.
- Automate useful security checks and work with engineers to make failures easy to investigate and resolve.
What we would look for
- Hands-on experience with CI/CD, infrastructure as code and operating applications in a cloud environment.
- An understanding of least privilege, software supply chain risks and the management of deployment credentials.
- A practical approach to automation, observability and documentation that helps teams maintain the systems you build.
The challenges
The goal is fast delivery with controls that provide real protection. You would balance developer feedback speed, reliable releases and the need to avoid introducing brittle security gates that teams cannot maintain.
Register interestEngineering / 05AI Engineer
Build agent discovery, evaluation and access analysis capabilities.
Scope and responsibilities
- Develop agent discovery, access analysis and evaluation features that connect models to carefully scoped tools.
- Build measurable test sets and compare model or workflow changes against reliability, cost and safety requirements.
- Collaborate with security researchers to turn identified failure modes into enforceable application boundaries.
What we would look for
- Experience shipping LLM applications, tool integrations or agent workflows beyond a standalone prompt demonstration.
- Strong software engineering skills and an ability to debug interactions between models, APIs and application state.
- An evaluation-driven mindset, including comfort analysing failures and documenting where automated results need human review.
The challenges
Useful agent behaviour must coexist with predictable access controls. You would work on reducing unreliable tool use and ambiguous outputs while keeping the system useful across changing models and real business workflows.
Register interestData / 06Data Analyst
Turn agent activity and access events into useful security insights.
Scope and responsibilities
- Define metrics for agent inventories, permission exposure and the outcomes of access reviews.
- Analyse activity patterns and communicate findings through clear dashboards, reports and focused investigations.
- Work with engineers to identify missing events, inconsistent definitions and data quality issues before they distort decisions.
What we would look for
- Strong SQL and experience analysing imperfect operational datasets with a suitable scripting or analysis toolkit.
- An ability to explain uncertainty, distinguish correlation from evidence of a cause, and choose useful baselines.
- Careful judgement around personal information, aggregation and the limits of what a dataset can support.
The challenges
Security telemetry is often incomplete and heavily shaped by the source system. The challenge is to produce useful comparisons and signals without presenting missing data or unusual behaviour as proof of an incident.
Register interestData / 07Data Engineer
Create reliable pipelines for agent inventories, tool events and access telemetry.
Scope and responsibilities
- Build ingestion and transformation pipelines for agent inventories, tool calls and data access events.
- Define stable schemas, validation rules and lineage so downstream teams can understand where records came from.
- Implement appropriate retention, access controls and recovery procedures for security-related datasets.
What we would look for
- Experience building dependable data pipelines and modelling events from multiple systems or APIs.
- Strong programming and SQL skills, with an understanding of retries, deduplication and schema evolution.
- An operational mindset: you can investigate pipeline failures, document trade-offs and make data quality visible.
The challenges
Integrations emit events with different semantics, delays and identifiers. You would need to preserve meaningful relationships while handling duplicate, missing or late events and avoiding unnecessary collection of sensitive content.
Register interestEngineering / 08Backend Engineer
Build the services behind agent discovery and least privilege workflows.
Scope and responsibilities
- Develop APIs and services for agent inventories, access reviews and the processing of security findings.
- Implement clear authentication and authorisation boundaries, including isolation between customer environments.
- Build dependable background processing and maintain useful observability, error handling and operational documentation.
What we would look for
- Experience delivering backend services and working with databases, API contracts and production debugging.
- A strong understanding of access control, input validation and the management of credentials or sensitive records.
- An ability to choose straightforward designs, review trade-offs and collaborate on interfaces that other teams can maintain.
The challenges
The service must represent complex permissions without weakening the boundaries it is meant to inspect. You would balance integration flexibility, reliable processing and secure handling of data from systems the product does not control.
Register interestEngineering / 09Frontend Engineer
Make complex agent permissions and security findings easy to understand.
Scope and responsibilities
- Build accessible interfaces for agent discovery, permission reviews and the investigation of security findings.
- Make complex relationships understandable through clear states, navigation, progressive disclosure and well-designed interactions.
- Work with designers and backend engineers on reliable forms, data loading, error recovery and responsive layouts.
What we would look for
- Strong HTML, CSS and JavaScript skills, with experience delivering maintainable product interfaces.
- Practical accessibility knowledge, including keyboard navigation, semantic markup and usable feedback for assistive technology.
- A careful approach to handling sensitive information in browser state, logs, links and client-side error reporting.
The challenges
Security interfaces can overwhelm users with unfamiliar entities and large amounts of detail. The challenge is to support accurate decisions without hiding important context, and to stay usable when data is incomplete or requests fail.
Register interestSecurity / 10Cloud Security Engineer
Assess how AI agents interact with cloud identities, workloads and data stores.
Scope and responsibilities
- Review cloud identities, policies and trust relationships used by agents and their supporting services.
- Identify excessive permissions, exposed credentials and unintended paths to sensitive workloads or data stores.
- Work with system owners to propose scoped controls and validate changes against legitimate business workflows.
What we would look for
- Practical cloud security experience and a detailed understanding of identity, permission boundaries and service-to-service access.
- An ability to read infrastructure configuration and explain how several policies combine into effective permissions.
- Experience documenting evidence and recommending changes that account for operational ownership and recovery needs.
The challenges
An apparently narrow identity can gain access indirectly through roles, resource policies or shared services. You would investigate these relationships while keeping proposed changes proportionate and avoiding disruption to essential workflows.
Register interestSecurity / 11Detection Engineer
Identify risky behaviour in agent and tool activity.
Scope and responsibilities
- Develop detection logic for suspicious tool use, unusual exports and changes in agent access patterns.
- Validate rules with representative examples, investigate false positives and document response guidance.
- Work with data and security teams to identify telemetry gaps and improve the quality of investigation context.
What we would look for
- Experience with security telemetry, detection development and the analysis of alerts or incident evidence.
- Strong query or scripting skills and an ability to reason about event timing, identity and missing signals.
- A disciplined approach to measuring rule performance and explaining what a detection can and cannot establish.
The challenges
Agent workflows can change rapidly, making normal behaviour difficult to define. You would aim for useful, explainable signals that help responders investigate rather than producing a high volume of alerts with little supporting context.
Register interestSecurity / 12AI Red Team Specialist
Test agent boundaries through explicitly scoped adversarial exercises.
Scope and responsibilities
- Design adversarial exercises for prompt injection, tool misuse and data exposure within an explicitly authorised scope.
- Create realistic test scenarios using controlled data and record the conditions needed to reproduce a failure.
- Partner with researchers and engineers to validate mitigations and communicate remaining risks after a retest.
What we would look for
- Experience with red teaming, application security testing or the evaluation of tool-using AI systems.
- An ability to trace an attack across instructions, model decisions, tool calls and downstream authorisation checks.
- Strong judgement around test boundaries, sensitive data and the communication of findings without sensationalising results.
The challenges
A convincing demonstration is not enough on its own. You would need to separate model behaviour from integration flaws, understand how representative a scenario is and show whether a proposed control meaningfully reduces the risk.
Register interestEngineering / 13MCP Integration Engineer
Connect agent tools and MCP servers to discovery and permission analysis workflows.
Scope and responsibilities
- Build integrations that reveal available tools, authentication methods and the access requirements of MCP servers.
- Develop robust handling for configuration differences, capability changes and failures in external services.
- Work with security and product teams to present integration scope and permission requirements clearly.
What we would look for
- Experience building API integrations and working with authentication protocols, service credentials and permission scopes.
- Strong debugging skills across client-server boundaries and familiarity with tool-based AI applications.
- An ability to document limitations, version assumptions and operational requirements for people maintaining an integration.
The challenges
An exposed tool name does not fully describe what it can do or which data it can reach. The challenge is to produce reliable access information across evolving servers without granting broader credentials than the task requires.
Register interestProduct / 14Product Designer
Help security teams see what agents can access and decide what to change.
Scope and responsibilities
- Research how teams discover agents, review permissions and decide which security findings to address first.
- Design workflows, prototypes and accessible interfaces that connect evidence with understandable actions.
- Collaborate with engineers and researchers to test designs against realistic tasks and refine confusing states.
What we would look for
- A portfolio demonstrating interaction design, thoughtful information hierarchy and decisions grounded in user needs.
- Experience making complex or technical workflows understandable without removing context that users need to act safely.
- Comfort working with ambiguity, asking precise questions and communicating trade-offs through clear prototypes or specifications.
The challenges
The product needs to serve people with different levels of security expertise. You would help make risk and uncertainty understandable, while avoiding interfaces that imply more confidence, coverage or automation than the underlying system provides.
Register interestProduct / 15Technical Security Writer
Make AI agent security understandable and actionable.
Scope and responsibilities
- Write practical assessment guidance, integration documentation and explanations of agent security findings.
- Work with specialists to verify claims, reproduce examples and keep documentation aligned with product behaviour.
- Develop consistent terminology and content structures that help users find relevant instructions and understand limitations.
What we would look for
- Excellent technical writing and editing skills, supported by samples that explain complex subjects clearly.
- An ability to investigate APIs, configurations and security concepts rather than relying only on second-hand descriptions.
- Careful source checking and an understanding of how to distinguish product capabilities, recommendations and research findings.
The challenges
Agent security terminology and product behaviour evolve quickly. You would keep material accurate and actionable without overstating protection, exposing sensitive details or allowing older instructions to drift away from the current implementation.
Register interestLET'S TALK
Tell us what you want to build.
Choose a role and share a little about your background. If your interests cross several areas, choose an open application.