Chief Product Officer — AI Safety Programme

The mandate

An AI-safety programme has expanded evaluation depth and safeguard coverage faster than it has designed products customers can select and fund. Engagements accumulate expensive model calls and expert review, while buyers struggle to understand which tier supports their deployment decision. The board has authorised an urgent new CPO seat to make safety capability usable and economically sustainable without diluting independent judgement.

Approximately 300 employees and material partners span product, research, evaluations, engineering, design, commercial and customer delivery from New York. The onsite CPO owns product strategy, portfolio, discovery, experience, product operations and lifecycle economics, reporting to the Group Chief Executive or designated executive committee sponsor. Safety disposition and technical release remain with independent authorities.

Product discovery begins with consequence. A model-selection comparison, pre-release hazard assessment, deployment gate and continuous monitor have different users, evidence and timeliness. Product managers will define the decision and risk if wrong before choosing tests. More evaluation is not automatically better when it arrives after the decision.

The portfolio needs intelligible tiers. Scope, model access, test families, expert review, turnaround, evidence and limitations should be explicit. The CPO will prevent commercial packaging from implying that a lower-cost tier certifies safety. Escalation to deeper work must follow findings and use, not a hidden upsell script.

Cost-to-serve will be designed into the product. Model calls, secure environments, dataset preparation, specialist review, customer integration and reporting should be attributed. Product leaders will examine marginal information gained from each costly step. Sampling or automation is acceptable only when evidence remains valid for the stated decision.

Customer inputs require readiness. Intended use, system boundary, users, model version, data context and acceptable outcome should be complete before work starts. The CPO will create a guided intake that exposes uncertainty rather than allowing teams to make assumptions to preserve turnaround.

Findings experience must support action. Severity, evidence, confidence, affected conditions and recommended response need clear presentation. Customers should be able to route, dispute and close findings while the original evidence remains immutable. A visually simple score cannot replace a difficult but necessary explanation.

Product changes will preserve evaluation comparability. New datasets, scoring and models can invalidate historical trends. Versioning and migration guidance should tell customers what can be compared. Product marketing and customer success will not present improvement where the measurement basis has changed.

The roadmap must balance emerging hazards with product reliability. Researchers may identify important new tests that require rapid inclusion; customers depend on stable APIs and reports. The CPO will establish experimental channels, acceptance gates and supported releases so novelty does not destabilise evidence already in use.

Safeguards that operate during deployment need clear failure modes. Filters, monitors, policy engines and intervention workflows may create latency, false positives or bypass risk. Product requirements will include fallback, human authority and observability. A safeguard cannot be sold as effective when customers cannot staff the response it generates.

Responsible product practice includes refusal. Requests seeking a predetermined rating, surveillance outside stated purpose or unauthorised model access will not enter the roadmap. Customer influence must not change independent findings. Exceptions require documented ethical, legal and safety review.

Adoption will be measured by decisions completed and risks addressed. Usage volume can reflect repeated work caused by unclear findings. Product analytics will track resolution, repeat evaluation, integration and customer effort alongside revenue and cost. The organisation will retire features that attract attention but do not improve safety action.

The new product organisation requires technical and domain credibility. Product managers will work directly with evaluators and customer control owners. Research participation should focus on judgement only they can provide; the product team must learn enough to avoid treating scientists as permanent translators.

What you will own

• AI-safety product and portfolio strategy.
• Decision-led discovery and evaluation tiers.
• Product cost-to-serve and evidence economics.
• Customer intake and findings experience.
• Measurement versioning and comparability.
• Deployment safeguards and response workflow.
• Ethical refusal and product-exception governance.
• Product organisation and capability.

The first 12 months

Within 30 days, classify current engagements by customer decision, expose cost and evidence gaps and stop any product claim that overstates the available tier. Recruit pivotal product leaders.

By month six, launch decision-specific evaluation tiers, guided intake and findings workflow with full cost attribution. Establish experimental and supported release paths with independent safety authorities.

At twelve months, improve gross contribution by 20 percentage points, reduce median time from finding to customer decision by 35% and cut avoidable expert rework by 40%. Ninety per cent of engagements must enter through a defined product tier, with no lower-cost offer presented as certification or independent disposition bypassed.

What the sponsor will examine

• Product scope beginning with consequential decisions.
• Lower-cost tiers keeping honest limitations.
• Cost changes preserving information validity.
• Findings designed for accountable action.
• Measurement changes not masquerading as improvement.
• Product refusing work that compromises independence.

The person

You bring 22–28 years in AI, cybersecurity, assurance or enterprise platform product leadership, including CPO authority. Your record includes technical product packaging, usage economics, evidence-heavy workflows, responsible product choices and teams operating between researchers and enterprise buyers.

Candidates must describe a product tier they narrowed because evidence could not support its promise and a costly workflow they redesigned without compromising the decision. This permanent role is onsite in New York with close research and customer engagement.

Compensation and terms

Base compensation is US$430,000–575,000 plus annual incentive and equity linked to evidence-led adoption, product contribution, customer action, responsible scope and team creation. The permanent onsite New York CPO reports to the Group Chief Executive or designated executive committee sponsor. This is an urgent newly authorised appointment.

Confidentiality

The programme, models, tests, customers, product economics, findings and safeguard designs remain confidential. Further disclosure requires conflicts clearance, suitable experience and signed confidentiality. Applicants must not contact AI developers, safety bodies, customers or product partners to infer the sponsor.