# Charity

> Source: https://aiwiki.ai/wiki/charity
> Updated: 2026-05-13
> Categories: AI Ethics, AI Tools & Products, ChatGPT
> From AI Wiki (https://aiwiki.ai), a free encyclopedia of artificial intelligence. Quote with attribution.

*See also: [Charity ChatGPT Plugins](/wiki/charity_chatgpt_plugins)*

**Charity, nonprofit, and humanitarian work** has become one of the most visible testbeds for applied [artificial intelligence](/wiki/artificial_intelligence). Aid agencies, conservation groups, accessibility startups, and grantmakers have moved from pilots to production deployments since roughly 2017, using [machine learning](/wiki/machine_learning), [computer vision](/wiki/computer_vision), and [large language models](/wiki/large_language_model) to target cash transfers, forecast famines, identify endangered animals from photos and sound, describe the world to blind users, and draft donor proposals. The most cited proof of concept is the 2020 Novissi program in Togo, where a team of researchers from UC Berkeley, [GiveDirectly](/wiki/givedirectly), Innovations for Poverty Action, and the Togolese government used satellite imagery and mobile phone metadata to route emergency COVID-19 cash to roughly 138,000 of the country's poorest informal workers. The work was later published in *[Nature](/wiki/nature)* in 2022.

Not every deployment is celebrated. The same technologies that make targeting more accurate also enable surveillance of refugees, biometric registration with weak consent, and algorithmic exclusion of people the training data misses. Practitioners and researchers at the International Committee of the Red Cross, the United Nations Office for the Coordination of Humanitarian Affairs (OCHA), Chatham House, and the OHCHR have all warned that humanitarian AI carries unusual risks because beneficiaries are usually unable to opt out.

## Overview

The phrase "AI for good" started circulating around 2017, when the [International Telecommunication Union](/wiki/international_telecommunication_union) (ITU) hosted the first AI for Good Global Summit in Geneva and [Microsoft](/wiki/microsoft) launched its AI for Earth program. Within a few years the ecosystem grew to include philanthropic AI labs at [Google](/wiki/google), Microsoft, and [DeepMind](/wiki/deepmind); grant programs from Google.org and Open Philanthropy; nonprofit-specific commercial products from [Salesforce](/wiki/salesforce), Bloomerang, and Givebutter; and discounted enterprise plans from [OpenAI](/wiki/openai) and [Anthropic](/wiki/anthropic).

The work clusters into a handful of recurring use cases:

| Domain | Typical AI technique | Representative project |
|---|---|---|
| Poverty targeting | Tabular ML on phone metadata, satellite CNNs | Togo Novissi (2020) |
| Humanitarian forecasting | Time-series and ensemble models | WFP HungerMap LIVE |
| Disaster response | Pre- and post-event satellite change detection | 510 Automated Damage Assessment |
| Conservation | Image and audio classifiers | Wildbook, Perch |
| Accessibility | Multimodal LLMs, OCR, object recognition | Be My AI, Seeing AI |
| Fundraising | Generative AI on CRM data | Salesforce Einstein, Bloomerang Penny |
| Mental health | Conversational simulators | Trevor Project Crisis Contact Simulator |

The rest of this article walks through each of these areas, then closes with documented concerns and limitations.

## Poverty targeting and cash transfers

### Togo Novissi (2020)

When [COVID-19](/wiki/covid_19) reached West Africa in early 2020, the Togolese government wanted to send unconditional cash to informal workers who had lost income during lockdown. Conventional means-testing was impossible: most beneficiaries had no bank account, no formal employer, and no recent household survey. The government launched Novissi ("solidarity" in Ewe) in April 2020, with the first phase reaching about 572,852 people in Greater Lomé through mobile money disbursements.

For the second phase, the government partnered with GiveDirectly, the Center for Effective Global Action (CEGA) at UC Berkeley, and Innovations for Poverty Action. Researchers Emily Aiken and Joshua Blumenstock built a two-stage targeting pipeline: high-resolution satellite imagery and household consumption surveys picked the 100 poorest cantons, and a classifier trained on mobile phone metadata picked the poorest individuals inside those cantons. Phone features included international call patterns, internet usage, and mobile money balances. The second phase paid roughly 57,000 new beneficiaries between November 2020 and March 2021. Total reach across both phases was about 138,000 people.

Aiken, Blumenstock, and co-authors published the methodology in *Nature* in March 2022 under the title "Machine learning and phone data can improve targeting of humanitarian aid." Relative to the geographic targeting options that Togo had been considering, the ML approach reduced errors of exclusion by 4 to 21 percent. Novissi won a Paris Peace Forum award and a 2022 SXSW Innovation Award.

### Famine Action Mechanism (2018)

The [World Bank](/wiki/world_bank), the [United Nations](/wiki/united_nations), and the International Committee of the Red Cross launched the Famine Action Mechanism (FAM) at the 2018 UN General Assembly on 23 September. Microsoft, Google, and Amazon Web Services built an ensemble called Artemis that ingests satellite imagery, market price data, weather, conflict reports, and social media to forecast worsening food security. The goal was to shift famine response from reactive payouts to anticipatory action, releasing pre-arranged funding before a crisis crosses the IPC famine threshold. Initial pilot countries included Somalia, South Sudan, Afghanistan, Niger, and Mali.

The FAM is one of several efforts that feed into the broader anticipatory action agenda at the UN's Central Emergency Response Fund (CERF), which began releasing funds based on pre-agreed model triggers in 2020.

## Humanitarian response and food security

### UN OCHA Centre for Humanitarian Data

OCHA opened the Centre for Humanitarian Data in The Hague in late 2017, with foundational funding from the Dutch Ministry of Foreign Affairs and the City of The Hague. The Centre runs four workstreams: data services (it hosts the Humanitarian Data Exchange, HDX), data literacy, data policy, and predictive analytics. The predictive analytics team partners with academic groups and operational agencies on models for cholera outbreaks, displacement flows, drought, and flooding. In March 2020 the Centre published a Peer Review Framework for Predictive Analytics in Humanitarian Response, a checklist used to vet whether a model is fit to trigger funding or operations.

### WFP HungerMap LIVE

The [World Food Programme](/wiki/world_food_programme) launched HungerMap LIVE in 2019 as a public dashboard for tracking food security across more than 90 countries. Where survey data is sparse the platform uses ML "nowcasts," supported by Google.org, that combine market prices, weather, conflict reports, and mobile vulnerability assessment data to estimate the share of a population that is food insecure in near real time. A 2024 update added AI-assisted forecasting in 16 hunger hotspot countries and a micronutrient adequacy metric developed with support from the Bill and Melinda Gates Foundation.

### UNICEF Magic Box

[UNICEF](/wiki/unicef) started Magic Box in 2014 to combine private-sector real-time data (call detail records from Telefonica, Vodafone, and others, plus Google search trends and Amadeus travel data) with public health and climate data. Early use cases included the Ebola response in 2014 and the Zika response in 2016. UNICEF later opened parts of the platform on GitHub.

### IFRC and Netherlands Red Cross 510

The Netherlands Red Cross runs a digital response team called 510 that deploys ML models in disasters. After the February 2023 [Türkiye](/wiki/turkey) and Syria earthquakes, 510 ran its Automated Damage Assessment (ADA) tool, a suite of deep learning models that compares pre- and post-event high-resolution satellite imagery to flag damaged buildings, helping the [International Federation of Red Cross and Red Crescent Societies](/wiki/ifrc) prioritize search and rescue. Multiple academic groups also published independent damage-detection results from the same event, using synthetic aperture radar change detection and deep ensembles on optical imagery.

## Conservation and biodiversity

### Wildbook and Wild Me

Wild Me, now part of Conservation X Labs, runs Wildbook, an open-source platform that uses computer vision to identify individual wild animals from photos submitted by researchers and citizen scientists. The flagship deployment is for whale sharks, identified by the unique spot pattern behind the gills; the database had more than 8,100 tagged individuals by the late 2010s. Wildbook later expanded to zebras, polar bears, ragged-tooth sharks, and others. A sibling platform, Flukebook, identifies humpback whales, sperm whales, and other cetaceans by the trailing edge and pigmentation of their flukes. As of the 2022 platform paper in *Mammalian Biology*, Flukebook hosted more than 2 million photos of over 52,000 identified individuals and ran 37 species-specific pipelines using algorithms such as HotSpotter, CurvRank v2, and PIE.

### Microsoft AI for Earth

Microsoft launched AI for Earth in July 2017. At the One Planet Summit in Paris that December, the company committed $50 million over five years for grants, Azure credits, and engineering support to organizations working on climate, agriculture, water, and biodiversity. Within the first six months it had funded more than 35 grantees in 10 countries. AI for Earth was later folded into Microsoft's broader AI for Good Research Lab.

### Google Perch and Cornell Merlin

Google Research released Perch in 2023, a bioacoustic classifier originally trained on bird recordings from Xeno-Canto and iNaturalist. Perch can predict species presence in field recordings and supports vector search so that a researcher can find more clips that sound like a single example. Perch has been downloaded over 250,000 times and has been adapted to marine mammals as well. A 2025 update from [Google DeepMind](/wiki/deepmind), Perch 2.0, covers nearly 15,000 species.

The Cornell Lab of Ornithology has run Merlin Bird ID since 2014. The photo-based identifier added a machine learning model around 2015. Sound ID, which identifies birds by song or call in real time on a phone, launched in June 2021. Sound ID is powered by Visipedia models trained on the Macaulay Library, which holds tens of millions of annotated photos and over a million sound clips uploaded by the eBird community.

### Other conservation tooling

Conservation groups also use passive acoustic monitoring with convolutional neural networks for at-risk species, automated camera trap pipelines such as Microsoft's MegaDetector, and species distribution models on remote sensing data. Many of these projects publish models openly on GitHub or Hugging Face, which has lowered the barrier for small conservation NGOs to adopt them.

## Accessibility, Be My AI, Seeing AI, Lookout

Digital accessibility is the area where mainstream AI products reached blind and low-vision users earliest and most directly.

### Be My Eyes and Be My AI

Be My Eyes is a Danish nonprofit that since 2015 has connected blind users with sighted volunteers over a live video call. On 14 March 2023, alongside the launch of [GPT-4](/wiki/gpt-4), Be My Eyes announced a Virtual Volunteer feature that uses GPT-4's image input to describe photos and answer follow-up questions. The feature was renamed Be My AI later in 2023 and reached general availability for iOS and Android by the end of the year. Users can take a photo of, say, the contents of their fridge, the laundry symbols on a clothing tag, or a stranger's facial expression, and ask follow-up questions in natural language.

### Seeing AI

Microsoft released Seeing AI for iOS on 12 July 2017 through the Microsoft Garage. The app reads short text on the fly, narrates printed documents with audio guidance for framing, identifies products by barcode, recognizes friends from saved photos, describes scenes and facial expressions, and reads currency. Seeing AI logged more than 3 million tasks in its first six months. The Android version launched in late 2023, and Microsoft has continued to add features such as multimodal scene narration powered by newer foundation models.

### Google Lookout

Google previewed Lookout at Google I/O 2018 and released it on Android in 2019. Lookout offers modes for currency identification, packaged food labels, document scanning, scene exploration, and image question-and-answer. It runs in more than 30 languages on devices with Android 6 and above.

### Aira and Envision

Aira, a paid service founded in 2014, connects users to human visual interpreters over a smartphone camera. Aira has added AI assistance from a virtual agent called Chloe for OCR and object recognition tasks that do not need a human. The Dutch company Envision sells the Envision Glasses, smart glasses based on the Google Glass Enterprise Edition 2 frame, which Envision announced in late 2020 and shipped through 2021. The glasses do real-time OCR in roughly 60 languages, recognize scenes and faces, and read handwritten text. In 2025 Envision announced an Ally Solos partnership for a lighter, phone-offloaded device at a much lower price.

| Tool | Organization | Year released | Approach |
|---|---|---|---|
| Be My Eyes (volunteer app) | Be My Eyes (nonprofit) | 2015 | Live video to sighted volunteers |
| Seeing AI | Microsoft | 2017 (iOS) | On-device and cloud vision models |
| Google Lookout | Google | 2018 preview, 2019 launch | On-device vision, modes |
| Aira | Aira Tech Corp | 2014 | Human agents plus Chloe AI |
| Envision Glasses | Envision (NL) | 2020 announcement | Google Glass with cloud AI |
| Be My AI | Be My Eyes + OpenAI | March 2023 beta, late 2023 GA | GPT-4 multimodal |

## Fundraising and nonprofit operations

Generative AI hit the fundraising stack in 2023 and 2024. The patterns are similar across vendors: write personalized emails and appeals from donor records, summarize meeting notes and grant applications, and surface the next-best action on a donor.

[Salesforce](/wiki/salesforce) embedded Einstein generative AI inside Nonprofit Cloud. Fundraising Gift Proposals drafts a personalized major-gift proposal grounded in the donor's giving history, program interests, and past interactions. Einstein Summaries condense grant applications, donor profiles, and program performance into one-page briefs. The features are gated to Nonprofit Cloud Enterprise and Unlimited Editions with the Sales or Service Einstein add-on.

Bloomerang, a CRM aimed at small and mid-size nonprofits, sells AI features under the name Penny, an action recommender that surfaces lapsed donors and suggests follow-ups, plus a content assistant for email drafting and "Smart Amounts" that personalize ask strings. The vendor claims a 55 percent lift in conversions on giving pages that use the recommended amounts.

Givebutter, a free fundraising platform, has rolled out AI-assisted campaign and email drafting since 2024. Its broader appeal to small nonprofits is that the underlying platform is free and monetized through optional tips from donors. Other CRMs (Blackbaud, Bonterra, DonorPerfect, Virtuous) have shipped comparable features.

At the foundation grant layer, [Candid](/wiki/candid) (the merger of Foundation Center and GuideStar) and Charity Navigator have both added AI-assisted search and summarization tools, although Charity Navigator's scoring system itself remains rule based as of this writing.

## AI for Good programs

A handful of corporate and intergovernmental programs anchor what people call the "AI for Good" ecosystem.

| Program | Organization | Launch year | Focus |
|---|---|---|---|
| AI for Good Global Summit | ITU (with 40+ UN partners) | 2017 | Annual Geneva summit, SDG alignment |
| AI for Earth | Microsoft | July 2017 | $50M over 5 years for climate and biodiversity |
| AI for Accessibility | Microsoft | May 2018 | Grants and engineering for disability access |
| AI for Humanitarian Action | Microsoft | September 2018 | $40M over 5 years for disaster and refugee response |
| AI for Health | Microsoft | January 2020 | $40M over 5 years for global health research |
| Google AI Impact Challenge | Google.org | 2018 | $25M to 20 nonprofits using AI |
| AI for Social Good (Google) | Google Research and Google.org | 2018 | Internal research and external partnerships |
| AI for Good Lab | Microsoft Research | 2018 | In-house research lab on humanitarian AI |
| Open Philanthropy AI safety RFPs | Open Philanthropy | 2014 onward | Technical AI safety and field building |
| Claude for Nonprofits | Anthropic + GivingTuesday | 2025 | Up to 75 percent discount on Team and Enterprise plans |
| ChatGPT Enterprise nonprofit discount | OpenAI | 2024 | Discounted plans for verified nonprofits |

The Google AI Impact Challenge, launched in late 2018, received over 2,600 applications from 119 countries and selected 20 grantees, including the Trevor Project (see below), Wild Me, Médecins Sans Frontières, and Quill.org. The challenge was paired with a custom nine-month accelerator run with DataKind.

Microsoft's family of "AI for..." programs sits under the AI for Good Research Lab, established in 2018 and led by Juan Lavista Ferres. The lab embeds data scientists with nonprofits and academic partners on specific projects rather than purely cutting checks.

Anthropic announced Claude for Nonprofits in late 2025 in partnership with GivingTuesday, offering discounts of up to 75 percent on Team and Enterprise plans, training through an AI Fluency for Nonprofits course, and partnerships with Bridgespan, Vera Solutions, Idealist Consulting, and Slalom for implementation help. Anthropic also announced a separate program with the Tipping Point Community in San Francisco to give 50 Bay Area nonprofits six months of free Enterprise access starting in late 2025.

## Crisis and mental-health support

### The Trevor Project

The Trevor Project provides crisis services for LGBTQ youth in the United States. In February 2021 the organization announced the Crisis Contact Simulator, an AI-powered chatbot built with support from Google.org that lets new counselors practice difficult conversations with simulated personas. The first persona, "Riley," was modeled on an anxious and depressed teen from North Carolina. A second persona, "Drew," representing a young adult facing harassment, was added in December 2021. The simulator was a finalist for the Fast Company Innovation by Design Awards in 2021. Within its first year the tool trained over 1,000 counselors. The Trevor Project still uses human counselors for live crisis contacts; the AI is restricted to training.

### Other deployments

Crisis Text Line, founded in 2013, has long used machine learning on incoming texts to triage high-risk conversations to the front of the queue. Childline, Samaritans, and several national suicide prevention services have piloted similar triage models. Use of generative AI to talk directly to people in crisis remains controversial after a high-profile incident in 2023 when the National Eating Disorders Association's chatbot Tessa was suspended for giving unsafe weight-loss advice.

## Education in low-income contexts

Khan Academy released Khanmigo, an AI tutor and teacher assistant built on GPT-4, on 14 March 2023 alongside the OpenAI GPT-4 launch. Khan Academy is a nonprofit, and Khanmigo is positioned as a way to bring tutoring to students who cannot afford a human tutor. By the 2023 to 2024 school year, over 40 US school districts and around 28,000 students and teachers were piloting it. Sal Khan has separately discussed partnerships with developing-country governments through Khan World School and Schoolhouse.world, although the deeper international rollouts are still gradual.

Other nonprofits in this space include Sikana, which produces practical video lessons translated by ML, and the African nonprofit Eneza Education, which has long used SMS-based learning bots in East Africa.

## Concerns and limitations

The research literature on humanitarian AI is unusually self-critical. A few of the documented concerns:

1. **Bias in training data.** Models trained mostly on Global North data can fail on populations they were not built for. A 2022 Chatham House report on refugee protection in the AI era noted that facial recognition systems have higher false-positive rates on women with darker skin and on people using assistive equipment, with direct consequences for asylum decisions and aid eligibility.
2. **Surveillance humanitarianism.** Biometric registration in refugee camps, originally framed as efficiency, has been criticized for creating durable databases that host governments or third parties can later use to track or deport beneficiaries. The Office of the UN High Commissioner for Human Rights (OHCHR) issued a 2025 brief calling for human-rights due diligence on digital technology in humanitarian contexts.
3. **Targeting failures.** The Togo Novissi work itself documented that ML targeting reduces exclusion errors but does not eliminate them. People without a mobile phone, women who use a male relative's phone, and informal workers in rural areas with low cell coverage are systematically harder to reach.
4. **Model drift in conflict zones.** Predictive analytics models trained on pre-crisis data lose accuracy when wars, displacement, or cash shortages change the underlying signal. OCHA's Peer Review Framework explicitly tests models for this kind of degradation.
5. **Donor dependence and lock-in.** Many of the AI tools used by aid agencies sit on cloud credits from Microsoft Azure, Google Cloud, or AWS. When credits expire, the model often does too. The Famine Action Mechanism's Artemis system has been criticized for this kind of fragility.
6. **Mission creep.** Once a humanitarian database exists, governments and police forces tend to ask for access. The International Committee of the Red Cross has published extensively on data protection in armed conflict precisely because of this risk.
7. **Generative AI hallucinations in fundraising.** Personalized donor outreach drafted by an LLM can invent donor history. Several Salesforce, Bloomerang, and Anthropic deployments include guardrails and require human review for that reason.

## See also

- [AI for good](/wiki/ai_for_good)
- [AI ethics](/wiki/ai_ethics)
- [AI in healthcare](/wiki/ai_in_healthcare)
- [AI in education](/wiki/ai_in_education)
- [AI safety](/wiki/ai_safety)
- [Be My Eyes](/wiki/be_my_eyes)
- [Computer vision](/wiki/computer_vision)
- [GPT-4](/wiki/gpt-4)
- [Machine learning](/wiki/machine_learning)
- [Open Philanthropy](/wiki/open_philanthropy)
- [Charity ChatGPT Plugins](/wiki/charity_chatgpt_plugins)

## References

1. Aiken, E., Bellue, S., Karlan, D., Udry, C., and Blumenstock, J. (2022). ["Machine learning and phone data can improve targeting of humanitarian aid."](https://www.nature.com/articles/s41586-022-04484-9) *Nature* 603, 864 to 870.
2. GiveDirectly. ["Innovations in Togo and the future of cash distribution."](https://www.givedirectly.org/mobileaid-work-in-togo/)
3. World Bank (2021). ["Prioritizing the poorest and most vulnerable in West Africa: Togo's Novissi platform."](https://www.worldbank.org/en/results/2021/04/13/prioritizing-the-poorest-and-most-vulnerable-in-west-africa-togo-s-novissi-platform-for-social-protection-uses-machine-l)
4. World Bank (2018). ["United Nations, World Bank, and Humanitarian Organizations Launch Innovative Partnership to End Famine."](https://www.worldbank.org/en/news/press-release/2018/09/23/united-nations-world-bank-humanitarian-organizations-launch-innovative-partnership-to-end-famine)
5. UN OCHA Centre for Humanitarian Data. ["The Centre's Predictive Analytics Focus."](https://centre.humdata.org/the-centres-predictive-analytics-focus-models-peer-review-community/)
6. WFP. ["HungerMap LIVE."](https://hungermap.wfp.org/)
7. WFP. ["WFP Releases HungerMap Live."](https://www.wfp.org/news/wfp-releases-hungermap-live-modernized-intelligence-platform-turns-data-global-hunger-early)
8. UNICEF Office of Innovation. ["Magic Box."](https://www.unicef.org/innovation/Magicbox)
9. Netherlands Red Cross 510 (2023). ["The Digital Response to the Earthquakes in Türkiye and Syria."](https://510.global/2023/06/the-digital-response-to-the-earthquakes-in-turkiye-and-syria/)
10. Wild Me. ["Wildbook."](https://www.wildme.org/wildbook.html)
11. Blair, P. et al. (2022). ["Flukebook: an open-source AI platform for cetacean photo identification."](https://link.springer.com/article/10.1007/s42991-021-00221-3) *Mammalian Biology*.
12. Microsoft (2017). ["AI for Earth can be a game-changer for our planet."](https://blogs.microsoft.com/on-the-issues/2017/12/11/ai-for-earth-can-be-a-game-changer-for-our-planet/)
13. MIT Technology Review (2017). ["Microsoft Announces $50 Million for Its AI for Earth Project."](https://www.technologyreview.com/2017/12/11/147154/microsoft-announces-50-million-for-its-ai-for-earth-project/)
14. Google DeepMind (2025). ["How AI is helping advance the science of bioacoustics to save endangered species."](https://deepmind.google/blog/how-ai-is-helping-advance-the-science-of-bioacoustics-to-save-endangered-species/)
15. Cornell Lab of Ornithology. ["The Magic of Merlin."](https://www.birds.cornell.edu/home/the-magic-of-merlin/)
16. Be My Eyes (2023). ["Introducing Be My AI for People who are Blind or Have Low Vision, Powered by OpenAI's GPT-4."](https://www.bemyeyes.com/news/introducing-be-my-ai-formerly-virtual-volunteer-for-people-who-are-blind-or-have-low-vision-powered-by-openais-gpt-4/)
17. Microsoft Accessibility Blog. ["Seeing AI: New Technology Research to Support the Blind and Visually Impaired Community."](https://blogs.microsoft.com/accessibility/seeing-ai/)
18. Google (2018). ["Lookout: an app to help blind and visually impaired people learn about their surroundings."](https://blog.google/outreach-initiatives/accessibility/lookout-app-help-blind-and-visually-impaired-people-learn-about-their-surroundings/)
19. Envision (2020). ["Envision Announces AI-Powered Smart Glasses For The Blind And Visually Impaired."](https://www.letsenvision.com/blog/envision-announces-ai-powered-smart-glasses-for-the-blind-and-visually-impaired)
20. Aira. ["Aira Visual Interpreting Service."](https://aira.io/)
21. Salesforce (2024). ["New Salesforce AI and Data Tools Help Nonprofits Engage with Stakeholders, Boost Fundraising, and Improve Program Efficiency."](https://www.salesforce.com/news/stories/ai-tools-for-nonprofits/)
22. Bloomerang. ["A Giving Platform Powered by AI for Nonprofits."](https://bloomerang.com/ai/)
23. ITU AI for Good. ["AI for Good Global Summit."](https://aiforgood.itu.int/)
24. Google.org (2019). ["Google AI Impact Challenge."](https://impactchallenge.withgoogle.com/ai2018)
25. The Trevor Project (2021). ["The Trevor Project Launches New AI Tool To Support Crisis Counselor Training."](https://www.prnewswire.com/news-releases/the-trevor-project-launches-new-ai-tool-to-support-crisis-counselor-training-301254382.html)
26. The Trevor Project (2021). ["The Trevor Project Expands Its AI Innovation with New Crisis Contact Simulator Persona."](https://www.prnewswire.com/news-releases/the-trevor-project-expands-its-ai-innovation-with-new-crisis-contact-simulator-persona-to-scale-counselor-training-301439302.html)
27. Khan Academy (2023). ["Harnessing GPT-4 so that all students benefit."](https://blog.khanacademy.org/harnessing-ai-so-that-all-students-benefit-a-nonprofit-approach-for-equal-access/)
28. Anthropic (2025). ["Introducing Claude for Nonprofits."](https://www.anthropic.com/news/claude-for-nonprofits)
29. Open Philanthropy. ["Request for Proposals: Technical AI Safety Research."](https://www.openphilanthropy.org/request-for-proposals-technical-ai-safety-research/)
30. Chatham House (2022). ["Refugee protection in the artificial intelligence era."](https://www.chathamhouse.org/sites/default/files/2022-09/2022-09-07-refugee-protection-artificial-intelligence-era-forster.pdf)
31. OHCHR (2025). ["AI: Human Rights Due Diligence for Digital Technology Use."](https://www.ohchr.org/sites/default/files/2025-09/ohchr-brief-ai.pdf)
32. ICRC International Review of the Red Cross. ["Harnessing the potential of artificial intelligence for humanitarian action."](https://international-review.icrc.org/articles/harnessing-the-potential-of-artificial-intelligence-for-humanitarian-action-919)