Conflicts of Interest Policy

This policy applies to our work as an independent AI evaluator. For our Financial Conflicts of Interest policy as an NIH grantee, see here.

As an independent third-party organization that assesses the capabilities of frontier models, we need to be objective and impartial in order to safeguard the credibility of our work. This policy is about the disclosure and management of potential conflicts of interest in order to ensure that our work and recommendations are free from bias, whether real or perceived, and is consistent with AEF-1’s guidance. For a full description of our principles for responsible model assessment, see this page.

A conflict of interest may arise when our team members or our organization have personal or financial relationships or interests that could affect, or appear to affect, the objectivity of our work. Examples are:

• Paid engagements with AI firms whose models we evaluate,
  
• Financial exposure to AI firms whose models we evaluate,
  
• Personal relationships (e.g. employment, relationships, group membership) with persons affiliated with AI firms we work with.

All staff and contractors on our AI team must disclose potential conflicts of interest when they begin working for SecureBio, and every 6 months thereafter. Disclosures are reviewed by the Executive Director and Director of AI in order to track and manage potential risks to the integrity of our work. When a conflict is identified, mitigation steps may include reassigning/recusing personnel, instituting independent review, or declining an engagement entirely.

Staff will be required to recuse themselves from engagements where an actual or perceived conflict of interest exists which cannot be effectively mitigated. Situations requiring recusal may include:

• Employment or consulting work with the assessed or audited entity within the last 12 months
  
• Financial holdings that could be affected¹ by the findings of our assessments or audits
  
• Close personal or family relationships with individuals in leadership roles at the assessed or audited entity

Recusals are documented, and alternative qualified personnel are assigned to preserve the integrity of the audit.

Principles governing our service/audit engagements

Our paid services cover safety evaluations and audits only, with compensation structured to cover the cost and time to deliver these services; in some cases, this may include credits for the compute needed to run the assessments. We only enter into engagements for which funding is not contingent on the results of our assessments and/or analyses. We disclose funding or the lack thereof in all public-facing reports.

Mitigation and oversight

Some conflicts can be managed rather than requiring recusal. Mitigation measures include steps like independent peer review, supervisory oversight, or transparency statements in audit reports. SecureBio leadership will maintain a record of all disclosures, reviews, and mitigation decisions, and periodically audits this process to ensure compliance.

External parties

When working with external consultants, subcontractors, or subject-matter experts who will be directly involved in frontier AI capability assessments or related evaluation work, we apply the same COI standards. All external contributors must complete a COI disclosure prior to engagement and agree in writing to abide by our independence policy. Conflicts that cannot be mitigated result in disqualification from the engagement.

Funding for projects

The organization endeavours to ensure that no more than 25% of its annual revenue is derived from services supplied to AI companies. AI firms, including those whose models we assess, provide us with free API credits that we use across our work, including in prerelease assessments. These credits are never contingent on the results of an assessment.

Footnotes

1. Any individual with a significant financial interest in a company whose model we are assessing must recuse from that engagement. At minimum, any direct equity in the evaluated entity held personally or by a spouse, spousal equivalent, or dependent. Excludes indirect holdings not managed directly (e.g. index or mutual funds) and support-only staff (logistics, infrastructure, tooling) overseen by qualified leadership with visibility into the relevant interests.