OnePointFive’s Policy for AI Use

Forward: Based on our experience with global carbon systems, project work dealing firsthand with AI innovations, and review of available nascent research, we believe that the scope of usage of AI systems and the composition of the infrastructure built to support them will determine the net impacts of the system as a whole.  Furthermore, we believe, if used well, AI can meaningfully accelerate progress on climate and sustainability. But that "if" matters. Much is left to be determined, including whether AI can meaningfully accelerate progress on climate and sustainability or contribute new, outsized impacts. AI systems have been shown to carry real environmental costs, and many available tools provide incomplete, inconsistent, or simply not transparent data about those costs. Given the fast pace of changes in AI, there is much to understand, while at the same time, the field of Sustainability and AI is nascent and also changing rapidly. This uncertainty must be understood and solved by teams of sustainably-minded and -skilled experts, otherwise, impacts will continue to grow unrestrictedly. What we can commit to internally is staying informed, keeping up with the latest developments, assessing available tools, and adopting them responsibly, thinking not just about how we use AI, but why, and how our usage can be shaped to protect our core values of environmental sustainability. This policy is our starting point, not our final answer: an imperfect but genuine attempt to use these tools conscientiously, while continually learning and leveraging our experiences to shape the future of the industry. Finally, in order to reduce bias, this entire policy has been deliberately developed without the use of AI. The perspectives in this policy are all human-generated by the OnePointFive team.

1. Introduction

1.1. Purpose 

Within what we can control at any given time, OnePointFive (OPF) commits to integrating sustainability into the procurement and use of Artificial Intelligence (AI) systems. This policy establishes guidelines accounting for AI’s environmental, social, and ethical impacts, aligning with our broader sustainability goals and values.

We aim to select, adopt, and use AI tools responsibly, critically considering emissions, sustainability impacts, ethical standards, long-term societal outcomes, and our potential for system-level influence.

1.2. Our Objectives & Why We Use AI

Our vision is that the climate crisis can be solved by deploying human ingenuity for the greatest good and to its fullest potential. And AI, a technology modeled on human intelligence, is one way we extend that ingenuity further. OPF has been using AI as a tool to help achieve its company mission of accelerating climate change solutions.

Our vision also guides how we approach AI; we do not believe AI should replace human ingenuity. Used deliberately, AI can be an accelerant: it lets impact-driven organizations, which are conventionally smaller in size, operate at the scale in which the climate crisis demands. Used carelessly, it can do the opposite - eroding the judgment, expertise, and rigor that make our work worth doing in the first place. Which outcome we get depends entirely on how intentionally we use it.

AI can accelerate our mission through various ways, including but not limited to:

  • Freeing up capacity for work that matters: By using AI for routine, repetitive, or time-intensive tasks, our team can focus its energy and bandwidth on work that requires a human touch, thought, expertise, and judgement.

  • Earning a seat at the table by leaning in: The standards, frameworks and best-practice approaches for measuring and reducing the environmental impact of  AI development and use are being written right now, and they will be written by the organizations actually doing the work. We cannot credibly shape guidance on responsible, climate-aware AI adoption from the sidelines. Engaging with these tools directly (and documenting what we learn) is one of the ways we can put ourselves  in a position to contribute. We treat our own adoption as a live internal case study: what we learn about using, measuring, and governing AI becomes expertise we can offer to clients, peers, and the wider industry.

  • Speeding up & improving our ability to deliver outcomes: AI expands our team's core competencies, access to knowledge, and problem-solving capabilities, helping us deliver high-quality work faster and take on more of the climate work that needs doing, thereby scaling up our impact.

  • Building future-proof solutions, approaches & tools: Developing AI-enabled tools to deliver our work; visualizing and tracking metrics and KPIs by integrating our company systems makes it easier to see areas of improvement, track our progress, and optimize our methods and strategies.

AI use also poses potential risks, including, but not limited to:

  • Environmental impacts & climate risks (e.g., water usage, energy consumption)

  • Data breaches and confidentiality infringement (e.g. client and company information)

  • Systemic impacts to the workforce (e.g., job displacement, reduced entry-level jobs)

  • Malfunctions (e.g., AI bias, hallucinations)

  • Operational risk (e.g., client relationship management, data quality, work product errors, regulatory & legal compliance)

  • Reputational risk to OPF as a company / brand (due to any of the above)

  • Potential conflict of interest with a more equitable world (e.g., cultural homogenization, widening wealth gap)

A conscious choice, not an inevitability. There is no consensus yet on the net climate impact of AI; we have seen evidence that AI can both accelerate climate solutions and that AI-driven productivity gains enable more emissions than they avoid,. We are not adopting AI because "it's here to stay"; we are adopting it because engaging thoughtfully and consciously will improve our ability to influence how AI is developed, used, and measured across our industry, and to drive towards more positive outcomes.

However, we remain in an exploratory posture: putting controls in place to mitigate risks where possible, investigating the ESG effects of our own usage, and reassessing honestly rather than defensively. We will keep monitoring this space and if the evidence changes our assessment of the tradeoffs, we will change our approach - including scaling back or stopping where warranted.

1.3. OPF’s Guiding Values Applied to AI

  • Commitment. Our commitment to quality work guides our internal scrutiny of AI output. We will hold ourselves accountable to a human-centered and practitioner-led method by keeping a ‘human-in-the-loop’ in the outputs of any AI tool. 

  • Collaboration. We value the team because together, we are greater than the sum of our parts. Through our AI usage, we will aim to collaborate better through saving time on administrative work, generating drafts for group iteration, and allowing team members to stay informed when absent (e.g., AI summaries). AI can become an anti-collaboration force, reducing touch points between team members and we will proactively work against that. 

  • Diversity. We believe that our differences generate strength, creativity and better solutions. AI can work against this: because models tend to produce similar, averaged answers, leaning on them too heavily can flatten the very range of perspectives we value. To counter this, we will make a conscious effort to seek one another’s opinions, to document our own thoughts first, and to not automatically default to the model-generated outputs of an AI tool, instead treating model-generated answers as one input among many.

  • Respect. We will be conscientious of any bias that AI can bring into our thought processes. In our AI usage, we will be open to prompt, direct, and “radical candor” feedback from our peers; and make sure we review AI outputs before sharing with other team members.

  • Trust. With trust comes the accountability for and honesty regarding our AI usage. We will not trust AI at face value. If we have any doubts or concerns, we will make them known to one another, both as creator and receiver of our team’s AI output. Due to the speed and newness of AI use, practising "radical candor" is essential when facing conflicts or issues with others and/or needing to provide real feedback. 

  • Transparency. We are open and clear about our thoughts and work, including whether our ideas originate from ourselves or AI tooling. We practice transparency in many forms, from documenting our sources, sharing our AI Policy publicly, and being open to raising questions or concerns that come up while engaging with AI systems. Furthermore, our client contracts will describe our AI use and confidentiality practices, so clients will be aware before engagement begins.

1.4. Definitions and Scope

This document applies to all employees (full-time and part-time) and contractors. For third-party partners of OPF, this policy applies on an awareness-and-acknowledgement basis.

  • When bringing on new contractors and partners, OPF includes a clause in their contracts confirming they have reviewed OPF’s AI Policy. OPF asks that partners complete their contributions and deliverables, whether independent or collaborative, with accountability, transparency, and in accordance with OPF’s AI standards. 

  • At contract signing, contractors and partners provide a short blurb of how they intend to use AI on the project, so we know what’s happening and can flag anything that conflicts with a client’s requirements or our Four Level Framework. 

  • Contractors and partners keep records of their AI use as the project progresses (specifically which tools they used and for what), so OPF can review them if questions arise. 

This policy governs the use of AI systems for OPF work, and any use involving OPF or client data, regardless of whether the tool is personal, free, company-provided, or client-provided.

2. Governance & Management

2.1. Governance Framework

This AI policy lays out a dedicated AI governance framework for OPF, assigning roles and responsibilities for the ethical, sustainable, and responsible deployment of AI, as well as the applicability of company values, security levels, responsible use, 3rd party management, and corrective action.

2.2. Roles & Bodies

  • AI Ethics & Policy Committee: OPF has an AI Governance committee consisting of employees from different levels of the company and departments, including senior leadership. The committee is responsible for setting OPF AI objectives, writing and revising this AI Policy, and remediating any issues of non-compliance. 

  • AI Policy Working Group: Responsible for researching, drafting, and defining general use policies for OPF AI usage. This working group includes an AI Policy Owner, who leads the policy development and supporting members.

  • Compliance Monitor: Responsible for auditing AI systems and updating AI related incident reports on a quarterly basis, and updating an audit log synthesizing findings to verify compliance with the policy. 

  • Open Door Policy: Anyone at OPF can raise topics or concerns to the Committee for discussion. There are two ways to do this. Employees can either raise it directly with their manager or bring it up on Slack by tagging the members of the AI Ethics & Policy Committee. 

2.3 Review & Improvement

The AI Ethics & Policy Committee will review this policy semi-annually given the pace of AI developments, and will make adjustments based on evolving best practices, technological advancements, and stakeholder feedback.

The Committee will review ongoing adherence to this policy quarterly. In these reviews, the Committee will go over the Compliance Monitor’s quarterly audit log, along with any documented incidents or raised concerns.

The following may also trigger an ad-hoc review:

  • A significant AI-related incident, internal or external 

  • Emergence of new, impactful AI technologies

  • Changes to relevant laws, regulations or standards

2.4 Preventive Controls 

We use three layers of guardrails to keep our AI use within policy. 

  • The first layer is guardrails integrating in our AI Systems. Every employee AI account is set up with the same instructions, which are rules entered into the tool's settings to align its outputs with this policy.

  • The second layer is risk management, which is the human-in-the-loop process to catch what the tools miss (See section 7). 

  • The third layer is audits. Through the quarterly internal audit process (See Sections 2.2 and 2.3 for additional information), we verify compliance with this policy and confirm that the first two layers are in place and working. 

3. Security & Privacy Measures

  • Use the right AI tool for the right data. We will treat data differently depending on the content. All employees will use the following four level framework to determine the appropriate security actions.

3.1. Four Level Framework

  • Data Classification System: We classify internal and client data into four tiers, based on their sensitivity and the harm that could result if the information was exposed. The tiers in the Data Classification System determines security requirements for what AI tool can be used.

    • Level 1 (Free tools) is for public data or personal use: Information that is already public or used for personal learnings, with no OPF or client content. 

    • Level 2 (Enterprise accounts) is for confidential business data: Most client work and most internal OPF work, with data protection guarantees. Most day-to-day work can use Level 2 business AI accounts.

    • Level 3 (Private hosting) is for restricted data: Data where the handling is constrained by contract, law or commercial sensitivity. Below are examples of when Level 3 would apply:

      • Contractual prohibition: client contract explicitly bans "third-party processing" or requires data to remain in client-controlled infrastructure

      • Data residency requirements: Data must stay in a specific country/region, as public SaaS platforms can't guarantee that

      • High-value IP: Proprietary algorithms, confidential data, or strategies where leakage would create direct commercial harm

      • Re-identification risk: Personal data that cannot be safely anonymized while maintaining utility

    • Level 4 (Maximum security) is for classified or regulated data: Data subject to the strictest legal requirements will only use AI systems that are fully offline (air-gapped).

      • Examples of when we would require this, including but not limited to: 

        • Classified government data

        • Highly regulated industries

        • Maximum security requirements

        • HIPAA-protected health data

  • Client and third-party transparency and alignment: All new client contracts signed after this Policy goes live will include a section describing our AI use methods and confidentiality methods. In addition, project teams will align at the beginning of projects or onboarding of new team members on what level of AI usage is allowed.

    • However, if you are unsure, before using Level 2 tools on a client project, the team should check whether the client’s contract prohibits third-party AI or data processing.

      • If the contract does not provide guidance, ask the project lead or the client if you're unsure. 

    • Don’t use Level 1 tools for client work unless you have fully anonymized the data for sandboxing and not for deliverables. 

    • Don’t put client deliverables or sensitive data in personal / free AI tools (Level 1). 

    • Keep records of which AI tools you used for client work.

3.2. Approved AI Tools Inventory 

The AI Ethics & Policy Committee maintains a list of approved AI systems to make sure the tools being used align with our values and objectives at OPF. All employees may only use tools on this list for internal OPF work or client work. Anyone wishing to use a new tool must request Committee approval before use. The Compliance Monitor verifies adherence to the list as part of the quarterly audit. 

Third-party partners and contractors are not restricted to the inventory list. Instead, per Section 1.4, they disclose their intended AI use at contract signing, and OPF reviews it for conflicts with client requirements and the Four Level Framework. This reflects the nature of their work; partners often specialize in domains outside our core practice and use tools for tasks that are unlikely to overlap with our day-to-day work. 

4. Sustainability of AI

As a climate-focused firm, we can’t advocate for emissions reduction without measuring the footprint of our own tools. This section discusses how we account for the impact of our AI use, our reduction efforts, and report it as part of our GHG emissions. The following factors are weighed heavily during the creation of OPF’s list of approved AI systems (Section 3.2)

  • Carbon Footprint Measurement & Reporting: Establish clear metrics and tracking systems for measuring the carbon footprint of our AI system design and use, incorporating it as part of our annual GHG accounting efforts and sustainability reporting. However, there are some known limitations with our AI carbon footprint, primarily coming from the lack of disclosures about location-specific energy use, server utilization, and carbon intensity by cloud and software providers. (See more information in our blog post about our first time measuring OPF’s Digital Emissions).

    • Our baseline digital footprint was 8.1 tCO₂e for the year 2024.

    • In 2024, OPF’s total emissions were 32.7 tCO₂e.

  • Renewable Energy: Whenever possible, select AI providers that are connected to data centers using renewable energy sources to power their AI systems, reducing reliance on non-renewable energy sources.

  • Social & Environmental Outcomes: When possible, we will align our selection of AI solutions with the Coalition for Sustainable AI’s Shared Vision and our organization’s broader environmental impact strategy, supporting goals such as resource conservation, reducing carbon emissions, and advancing social equity.

  • Emissions Reduction Solutions: Prioritize AI applications that directly contribute to emissions reduction goals and climate impact assessment capabilities. When choosing among different AI models, use publicly available tools to compare energy efficiency for various tasks and select the most sustainable option.

  • AI Vendor & Model Selection: When selecting AI vendors and models, we will evaluate their sustainability practices, including energy consumption, emissions, and overall environmental impact. We will prioritize partnerships or business with vendors who share our commitment to sustainability and ethical AI.

  • Efficient AI Use: Our team follows best practices For sustainable digital behavior to improve the efficiency of AI usage and receive training on this topic (e.g., starting new chats to keep threads short, writing concise prompts, and pasting text instead of using screenshots where feasible)

5. Supporting & Practicing Responsible AI

  • Knowledge Sharing & Collaboration: Establish protocols for sharing our AI sustainability practices or AI applications with the broader sustainability community while protecting sensitive methodologies. Communicate with sustainability peers and organizations so that we make the best use of AI tools while being transparent about our impacts.

  • Standards & Regulations: Actively participate in initiatives that can inform the development of industry standards for sustainable AI, leveraging our expertise in climate impact assessment. In addition to our contributions, we will track emerging standards and frameworks and apply the latest to how we use, measure, and mitigate the environmental impact of AI.

  • Partnerships for Sustainability: Choose to collaborate with stakeholders, including sustainability-focused organizations and research institutes, that are aligned with our sustainable AI policies and that choose to contribute to environmental goals.

6. Ethics 

  • Bias Mitigation: We acknowledge that AI models can carry inherent biases from training data. We select and prompt AI systems with fairness and inclusivity in mind, so that AI decision-making processes do not disproportionately impact marginalized or vulnerable groups. We will take proactive steps to identify and reduce biases when creating prompts and designing systems at every stage: 

    • We prompt models to take a representative view of diverse opinions when forming an output.

    • We have a diverse team from a range of geographies, cultures, experiences and professional backgrounds. This range of perspectives helps us catch bias that one perspective might miss. 

    • We review AI outputs with a lens for bias before use. When bias is a particular concern, we first bring in other team members; a co-worker or the wider team will review the output, drawing on the range of perspectives at OPF. If questions remain after human discussion, we run the same prompt through multiple models and compare the outputs for divergence, keeping this heavier use of AI a last resort rather than a default.

    • We are deliberate about where we choose not to use AI, particularly in cases where the work requires an objective and/or subjectively human perspective. For example, we did not use AI to create this policy.

  • Align with OPF principles: We will align our AI uses, reporting and disclosures with our mission, vision and objectives (see Section 1.2 and 1.3).

  • Data Ethics: We practice responsible data practices, including obtaining informed consent from individuals whose personal data we process, adding an AI section to our contracts with clients, safeguarding data privacy, and respecting the rights of individuals.

7. Human Review

  • A human will review every external company output (e.g., documents, deliverables, social media posts, presentations), at a minimum - once, verifying the information is accurate, free of hallucinations and preferably aiming for a second opinion. 

    • This is to align with our human-centered and practitioner-led method by keeping a ‘human-in-the-loop’ for the output of AI tooling and deliverables. 

  • A human will seek AI review when appropriate for human developed outputs, checking for errors, gaps and inconsistencies. This is a supplement for human review, not a replacement. Furthermore, the final review will always be completed by a human.

8. Legal Compliance 

  • Regulatory Compliance: We will comply with applicable local and global regulations to which we are subject, regarding emissions, data privacy, and AI ethics, such as the EU AI Act.

9. Noncompliance and Corrective Action

Below are the steps to follow if anyone violates any element of this policy. 

  • Stop immediately. Do not attempt to fix the problem on your own and stop using the tool. 

  • Notify the AI Policy and Ethics Committee using Slack

    • The Committee will conduct an impact assessment 

    • The Committee will determine next steps.

    • Next steps will include notifying affected parties.

    • The Committee will also do a root-cause analysis where warranted

    • We will treat honest mistakes reported promptly differently than deliberate misuse. The Committee will determine corrective action for deliberate, irresponsible AI use on a case-by-case basis. 

  • Document the incident as a memo for full review by the committee. 

Errors: Users of AI systems are responsible for overseeing the quality of their outputs. 

  • Errors that reach our work through AI (e.g., hallucinations, rework, or poor judgments based on unverified outputs) are the responsibility of the person who used the tool, just as if the error were their own. This individual accountability is separate from any conversation about continuing or discontinuing the usage of specific tools.

  • Humans must be able to pause, override or shut down AI systems when necessary. Where critical services rely on AI systems, we will have a plan to use manual alternatives in case the system fails or we need to take it offline.

Open Reporting: Any employee may raise noncompliance concerns directly to the committee for review. We will never retaliate against anyone who raises a concern in good faith. 

¹ AI Systems: Any software, tool, or service that uses Machine Learning, large language models or small language models, or any other artificial intelligence techniques, such as generative AI assistants like chatbots and AI features embedded inside everyday software.

² Stern, N., Romani, M., Pierfederici, R. et al. Green and intelligent: the role of AI in the climate transition. npj Clim. Action 4, 56 (2025). https://doi.org/10.1038/s44168-025-00252-3

³ Alpine, W., Geldner, N., Alpine, H. et al. AI-driven productivity gains enable more CO₂ emissions than they avoid in a global energy–economy model. npj Clim. Action5, 71 (2026). https://doi.org/10.1038/s44168-026-00411-0

⁴ Third-Party Partners: External individuals or organizations that contribute to OPF projects or handle OPF or client data, such as contractors, subcontracted firms, and co-delivery partners.