CTO – Product and Engineering — AI Safety Programme

The mandate

An AI-safety programme has proved technical demand through bespoke evaluations, red-team exercises and deployment controls, but its engineering model still depends on senior researchers assembling each engagement. Enterprise customers now expect repeatability, service commitments and integration support. The serving CTO will leave through planned succession; the replacement must commercialise capability without turning safety judgement into a feature factory.

Approximately 500 employees and material partners span research engineering, product development, security, evaluations, infrastructure, customer integration and technical operations. The hybrid New York CTO owns product engineering, architecture, reliability, technical delivery, developer experience and engineering talent, reporting to the Group Chief Executive or designated executive committee sponsor. Independent safety and release authorities retain their decision rights.

The first architecture task is to separate reusable platform from case-specific judgement. Evaluation orchestration, model access, test environments, evidence capture and reporting should be dependable services. Hazard definitions, scenario design and disposition may remain expert-led. The CTO will prevent standardisation from stripping context that determines whether a result is meaningful.

Commercial service levels require explicit boundaries. Model providers can change interfaces, weights or safety layers without notice; sensitive customer environments may restrict observability. Contracts and technical designs must state availability, latency, evidence and dependency assumptions. Engineering will not promise deterministic results where upstream behaviour remains outside control.

Evaluation integrity is a product property. Dataset versions, prompts, tools, scoring, reviewer intervention and environment configuration must be traceable. Re-running a test should reproduce the method and explain legitimate variance. The CTO will establish provenance and change controls that withstand customer, regulator and internal challenge.

Security architecture must protect unreleased models, vulnerability findings and adversarial methods. Isolation, privileged access, export, retention and customer tenancy need review at design time. Developer convenience cannot lead to safety findings appearing in general telemetry or shared support tools.

The engineering roadmap will reduce manual assembly while preserving escalation. Common connectors, controlled templates and automated evidence packaging can improve throughput; severe or novel findings still require qualified human review. The CTO will measure where automation introduces false confidence or shifts work into hidden verification queues.

Reliability will be tested through degraded conditions. Provider throttling, tool failure, corrupted artefacts and incomplete logs must yield safe, visible states. A customer should know when an evaluation is invalid or partial. The platform must never manufacture a complete-looking report from missing execution evidence.

Cost engineering matters at commercial scale. Compute, model calls, secure environments, expert review and storage should be attributable by product and engagement. The CTO will partner with product and finance on budgets and design alternatives, ensuring that cheaper sampling does not invalidate coverage claims.

Customer integration needs a defined engineering path. Identity, model endpoints, data boundaries, deployment workflows and findings management should use supported patterns. Strategic exceptions require architecture ownership and an exit plan. The most influential customer cannot quietly become the platform blueprint.

Release governance will connect code change to safety consequence. Tests, independent review, rollback and customer communication should reflect risk, not one uniform process. Emergency fixes require retrospective evidence and expiry. Engineering velocity will be assessed through stable, accepted capability rather than deployment count.

The succession must retain technical credibility. The outgoing CTO's architectural exceptions, provider relationships and unresolved reliability concerns need a tested handover. Engineering leaders below the role will gain clear domains and exposure to board decisions, reducing dependence on one executive's informal judgement.

What you will own

• AI-safety product engineering and architecture.
• Evaluation orchestration and evidence provenance.
• Reliability, security and safe degradation.
• Commercial service boundaries and integrations.
• Compute and expert-review cost engineering.
• Risk-based release and exception governance.
• Technical partners and platform dependencies.
• Engineering organisation, transition and succession.

The first 12 months

Within 45 days, secure the incumbent handover, map platform versus expert work and identify any commercial promise unsupported by engineering evidence. Stabilise critical provider and customer dependencies.

By month six, launch versioned evaluation orchestration, supported integration patterns and cost attribution. Exercise degraded modes with customer-facing teams and clarify independent release gates.

At twelve months, reduce bespoke engineering time per evaluation by 40%, improve critical-service availability to 99.95% and cut unplanned compute variance by 30%. Every customer report must carry reproducible method lineage, while severe findings continue to receive authorised human disposition and no security boundary is weakened.

What the sponsor will examine

• Reuse achieved without erasing hazard context.
• Evaluation evidence reproducible and versioned.
• Incomplete execution producing an honest state.
• Cost reductions preserving coverage validity.
• Customer exceptions governed and removable.
• Engineering leadership durable beyond succession.

The person

You bring 18–22 years in AI, developer infrastructure, security engineering or high-assurance software, including CTO or large engineering authority. Your record includes commercialising research capability, operating sensitive multi-tenant services and defending technical evidence with enterprise customers.

Candidates must show a product release they stopped because evaluation lineage was inadequate and a platform economy achieved without narrowing safety coverage. The role is hybrid in New York with significant presence alongside engineering, research and strategic customers.

Compensation and terms

Base compensation is US$430,000–575,000 plus annual incentive and equity linked to reliable commercialisation, engineering economics, evidence integrity, security and succession. This permanent hybrid New York CTO reports to the Group Chief Executive or designated executive committee sponsor. Planned replacement allows a structured incumbent transition.

Confidentiality

The organisation, models, customers, evaluations, vulnerabilities, engineering estate and succession details are confidential. Further disclosure requires conflicts clearance, appropriate credentials and signed confidentiality. Candidates must not approach AI laboratories, customers, safety organisations or infrastructure providers to determine the sponsor.