Ethical Tech CoLab

🛰 We put this month's featured tool to work

HASTE (see AI for Good, below) speaks directly to a gap in our Mariupol Corridor Severity Model, the daily civilian-danger index we built for the 2022 siege. That model's infrastructure-damage component leans on UNOSAT satellite assessments, which were published on only five dates across the 77-day siege and then joined by straight lines in between, meaning a stretch with no new imagery could look calmer than the city actually was.

HASTE's approach, described under AI for Good below, is exactly what a denser, more honest damage curve needs. We are prototyping HASTE-style assessments to fill the days UNOSAT never imaged, so that a gap in published reports can no longer be mistaken for a gap in destruction.

🌎 Microsoft's AI for Good Lab open-sources HASTE

One of our favorite releases this month is HASTE (High-speed Assessment and Satellite Tracking for Emergencies), an open-source platform from Microsoft's AI for Good Lab that helps emergency responders map building damage after disasters using satellite imagery. Instead of requiring perfectly matched "before" and "after" images or large training datasets, HASTE lets analysts label a small number of damaged buildings and rapidly generate damage assessments across entire affected areas. The paper landed on arXiv on July 13 and the code is public.

Why Ethical Tech CoLab loves it

• Open source: github.com/microsoft/haste
• Humanitarian impact
• Practical AI deployment
• Accessible to non-ML specialists

📄 Paper: arxiv.org/abs/2607.11838

🧬 Microsoft invests $60 million in AI for scientific discovery

On July 22, Microsoft announced a $60 million commitment to support the U.S. Department of Energy's Genesis Mission through a new coordination hub called SPARK (Scientific Partnership Advancing Research and Knowledge): roughly $40 million in Azure compute credits plus $20 million in engineering support connecting national laboratories, scientific data, and advanced AI infrastructure.

This is a reminder that some of AI's most transformative applications may come not from consumer products, but from accelerating science itself.

🔗 Microsoft announcement

🌍 AI for Good Global Summit 2026

The ITU AI for Good Global Summit (July 7 to 10, Geneva) once again demonstrated how AI is moving beyond prototypes into real deployments. Highlights included:

• the Robotics for Good Youth Challenge grand finale
• the Innovation Factory startup competition
• machine learning challenges spanning space AI computing and edge AI
• public sector AI pilots focused on health, agriculture, and climate resilience
• the AI for Good Impact Awards recognizing projects delivering measurable societal value

Geneva also hosted the UN's first Global Dialogue on AI Governance the same week (July 6 and 7), where Secretary-General Guterres pressed for urgent international controls: "Machines can inform, but humans must decide, and answer." Outcomes included a Child Safety Pledge and calls for mandatory human oversight of AI decisions in justice, healthcare, and policing.

🔗 AI for Good · UN News on the Global Dialogue

🌱 Communities owning their own AI data

Microsoft researchers introduced the Community Library Creator, a tool that lets communities build image and video libraries they own and control, deciding how their data is shared and requesting removal at any time. The pilot centers on underrepresented communities, with early work alongside disability communities.

This is an encouraging direction for anyone working on community data governance, humanitarian contexts, and digital public goods.

🔗 Microsoft Signal, July 22

EU AI Act transparency rules bite on August 2. The EU deferred the AI Act's high risk regime to late 2027 and 2028, but the Article 50 transparency obligations still take effect August 2, 2026: chatbot disclosure, synthetic media marking, deepfake labeling, and biometric categorization notices. Directly relevant to anyone working on displacement contexts and information integrity. Analysis

Illinois becomes the first US state to mandate independent frontier AI audits. Governor Pritzker signed SB 315 in early July, requiring annual third party safety audits of the largest frontier models. External assurance is moving from best practice to legal requirement. Announcement

UNHCR warns that AI-amplified misinformation is harming refugees. In a July 7 briefing, UNHCR reported that 93 percent of surveyed staff have witnessed misinformation or disinformation affecting operations, with deepfakes of humanitarian workers a growing tactic. The same week, the International Rescue Committee called on the tech industry to fund responsible AI for the world's 118 million displaced people. UNHCR · IRC

⭐ Sayari

What they do: Sayari builds graph intelligence software that maps corporate ownership, supply chains, sanctions exposure, and financial networks. Governments, financial institutions, and investigative teams use the platform for risk analysis, due diligence, and financial crime work.

Why we're watching

AI-assisted investigationsSupply chain transparencyAnti-traffickingFinancial crimeNational security

Hiring note: current openings skew toward engineering, sales, and delivery rather than analyst positions. Analyst roles appear periodically: sayari.com/careers

⭐ Tavily

Tavily has become one of the leading search APIs for AI agents, enabling large language models to retrieve reliable web information with citations. It has quickly gained traction among developers building research assistants and autonomous workflows.

AI infrastructureDeveloper-first toolsOpen-source adoptionRAG and research

Hiring now: Sales Development Representative, Austin · Forward Deployed Engineer, NYC

⭐ FutureHouse

FutureHouse is a San Francisco nonprofit building AI systems designed specifically for scientific discovery, helping researchers navigate literature, generate hypotheses, and accelerate biomedical research. Its Kosmos system is billed as its first "AI Scientist," and a commercial spinout, Edison Scientific, launched late last year.

If successful, platforms like these could significantly shorten the path from scientific question to experimental insight.

🔗 futurehouse.org

NotebookLM

Google's NotebookLM continues to stand out as one of the most useful AI tools for researchers, and its June update added the ability to build your source library from chat, plus editable outputs in PDF, Word, CSV, and slides. Coverage

• reading reports
• synthesizing PDFs
• generating study guides
• podcast style audio summaries
• literature reviews

Our rating: ⭐⭐⭐⭐⭐

HASTE: A Platform for Rapid Post-Disaster Building Damage Assessment

The platform is covered above; what the paper adds is the measurement behind it. On the xBD benchmark, one per cent of the labels scores 0.84 on area under the ROC curve against 0.88 for a fully supervised model, and the fast route passes that baseline at ten per cent, reaching 0.91. Read precisely, one per cent buys most of the accuracy rather than equal accuracy, hours sooner and without an ML engineer. Every figure is the developers' own and none has been independently reproduced.

We read the source code alongside the paper and wrote a plain-language report on what each setting does, what the numbers establish, and where the answer is constrained by things the software does not control. Read our report →

📄 Read on arXiv

Dwarkesh Podcast: Terence Tao on Kepler, Newton, and the true nature of mathematical discovery (March 2026). A thoughtful conversation on how AI is changing scientific and mathematical discovery, not by replacing researchers but by expanding what they can explore. Listen

Your Undivided Attention: Can AI Be Built in Service of Life? A conversation with Krista Tippett (July 16, 2026). Tristan Harris and the On Being host on trust and resisting "the spell of inevitability." Listen

WarGames (1983). A teenager dials into what he takes for a games company and starts a round of Global Thermonuclear War with the system holding the launch codes. Four decades on it is still the sharpest popular treatment of a problem we have not solved: a machine pursuing the objective it was given, and nobody in the loop able to tell a simulation from the real thing. Its famous closing line, that the only winning move is not to play, is really a lesson about how you specify a goal.

We built on it. Our War Games demo is a terminal thriller in the shape of the 1983 film, reframed around a modern AI agent, where the only winning move is to understand the machine. It ships with a Monte Carlo simulation harness and a written case study behind the fiction. Play the demo → · Repo

Also worth a look: The AI Doc: Or How I Became an Apocaloptimist (2026), from the director of Navalny, premiered at Sundance and reached theaters in March.

Atlas of AI, by Kate Crawford. Although published several years ago, it remains one of the strongest introductions to the social, environmental, and political dimensions of AI systems.

This section rotates monthly. Each edition features one mentor: biography, expertise, current role, recommended resources, office hours, and LinkedIn.

Prof. Yorke E. Rhodes III

Yorke works at the intersection of blockchain, artificial intelligence, and ethical systems design, applying emerging technology to real-world challenges from forced-labor mitigation to responsible AI deployment. As Director of Traceability he drives work on transparency and trust across global supply chains, and as an educator he shapes the next generation of ethical technologists through hands-on learning and thought leadership.

🔗 LinkedIn

Mentor tracks we are recruiting for: AI governance, humanitarian technology, blockchain for social impact, startup founders, and UN innovation specialists. Want to mentor? Reply to this email.