AI Policy

1. Purpose and Scope

The purpose of this policy is to enable Butterfly Ventures employees to use artificial intelligence (AI) tools productively while protecting confidential information, maintaining professional standards, and ensuring that accountability always remains with a human decision-maker.  AI is a tool that assists our work but does not replace judgement, responsibility, legal review, investment decision-making, or professional accountability. Every action performed using AI remains the responsibility of the employee.

The policy applies to all employees, partners, contractors, advisors, and interns, and covers all firm activities, including deal sourcing, investment screening and diligence, portfolio support, LP relations and fundraising, fund administration and operations, marketing and communications, and recruiting and HR.

2. Core Principles

Human Accountability

  • Employees remain responsible for all work products created with AI assistance
  • AI use is not a justification for inaccurate, misleading, incomplete, or inappropriate work
  • Important decisions must always involve human review and approval

Confidentiality First

  • Access to data must be limited to what is necessary for the task
  • Confidential LP, portfolio company, founder, employee, and business information may only be processed through approved and sanctioned tools and connectors
  • Users must follow company security requirements when enabling AI integrations and connectors

Transparency

  • Employees should be aware when material is substantially created or analyzed using AI
  • AI-produced outputs must be treated as drafts or recommendations until validated by a human reviewer

3. Approved Use of AI

AI tools may be used for:

  • Research and market analysis
  • Summarizing documents and meetings
  • Preparing first drafts of reports, presentations, and communications
  • Organizing and analyzing structured data
  • Drafting internal documentation
  • Supporting due diligence and investment analysis
  • Automating low-risk administrative tasks subject to review

Employees should begin with read-only and low-risk use cases before using AI in workflows that modify systems or records.

4. Data Protection and Confidentiality

The following activities require heightened caution and human verification:

  • Creating or modifying CRM records
  • Producing investment recommendations
  • Portfolio company assessments
  • Legal agreements and term sheets
  • Financial models and calculations
  • LP-facing communications
  • External communications issued in Butterfly’s name

Where practical, AI-generated changes should be reviewed through a staged approval process before execution.

5. Prohibited Uses

The following uses are prohibited:

No Autonomous External Decisions

  • AI may not independently send LP-facing, founder-facing, investor-facing, or legally binding communications
  • AI may not approve investments, contracts, payments, or legal commitments

No Credentials or Sensitive Authorizations

  • AI tools must never receive passwords, private keys, authentication credentials, or signing authority
  • AI may not sign, purchase, or make commitments on behalf of Butterfly Ventures

No Delegation of Personnel Decisions

  • AI may not make hiring, performance, promotion, compensation, disciplinary, or reference decisions
  • AI may support administrative preparation but not decision-making regarding people

No Unapproved Data Sharing

  • Confidential information may not be uploaded to consumer AI tools without explicit approval and appropriate safeguards.
  • Users must not connect data sources that are unnecessary for the task

6. Working Practices

When using AI for significant tasks, employees should:

  1. Clearly define the goal and context
  2. Ask the AI to produce a plan before acting
  3. Specify boundaries and approval requirements
  4. Limit access to relevant files and systems only
  5. Review outputs critically
  6. Verify facts, calculations, assumptions, and conclusions
  7. Maintain records of significant AI-assisted actions
  8. Use staged execution when updating systems

7. AI Risks

AI systems can be manipulated by instructions embedded within documents, websites, presentations, emails, or other materials being analyzed.

Employees must:

  • Treat AI outputs as potentially influenced by malicious or misleading instructions
  • Verify important outputs independently
  • Never allow AI systems to execute actions solely based on instructions found within analysed materials

Prompt injection risk shall be considered whenever AI is given access to external content.

8. Approved AI Tools

Butterfly Ventures may approve multiple AI tools where they provide business value and meet security requirements and do not limit itself to a single AI provider where other approved tools offer material advantages.  Approval decisions should consider:

  • Data protection
  • Security controls
  • Compliance obligations
  • Auditability
  • Quality and practical usefulness

9. AI in Investment Activities

Sourcing and Screening

AI may help identify relevant companies, sectors, themes, and opportunities. AI must not evaluate founders using characteristics or proxies such as gender, age, ethnicity, religion, nationality, sexual orientation, disability, appearance or other protected characteristics.

Due Diligence

For potential portfolio companies, due diligence should consider:

  • Data provenance
  • Data licensing
  • Security controls
  • Intellectual property ownership
  • Regulatory exposure
  • Model risks
  • Compliance with applicable regulation

10. AI in Recruiting and HR

Hiring is considered a high-risk AI use case. The following are not permitted:

  • Automatic rejection of candidates
  • Emotion recognition
  • Personality inference
  • “Culture fit” scoring
  • Behavioural inference from video or voice
  • Inferring protected characteristics
  • Scraping personal social media accounts for profiling
  • Using candidate data to train AI models
  • Undisclosed AI interviews or synthetic interviewers

No candidate may be rejected solely based on an AI recommendation. A manager must review every adverse hiring outcome.

11. Review

The policy will be reviewed periodically as AI capabilities, regulations, and business practices evolve, or following the introduction of new AI tools.

Quick Rules for Everyone

  1. Use approved tools only.
  2. Never put confidential data into unapproved AI systems.
  3. Check NDA restrictions before using AI on third-party materials.
  4. Verify facts, figures, calculations, and citations.
  5. AI never makes the final decision on an investment, hiring outcome, or LP communication.
  6. External communication always requires human review.
  7. If unsure, ask before you paste.
  8. Responsibility always remains with humans.

Last updated 5 October 2026.