AI safety with the subaltern in mind
Nairobi, Kenya
Joined June 2022
- Tweets452
- Following58
- Followers1.8K
- Likes457
ILINA Program retweeted
I talk a bit about risk thresholds and why they matter, especially when we're conducting safety evals but even more so when we're setting policies.
Our Research Scholar @gathoni_ireri was recently interviewed as part of a Rest of World story on how African countries are navigating AI safety. She highlighted the importance of context-specific AI safety evaluation. Read the story and her contribution: restofworld.org/2026/african…
ILINA Program retweeted
California's TFAIA and New York's RAISE Act both include whistleblower provisions for AI workers. Which better protects the people building frontier AI?
Neither, categorically. It depends. Our new comparative guide breaks it down.
What the guide finds:
📌 New York offers broader substantive coverage: contractors and former employees are explicitly protected, and any worker can report a "substantial and specific danger to public health or safety".
📌 California offers a more worker-friendly burden of proof: a "contributing factor" test shifts the burden to the employer to prove it would have acted the same way anyway.
📌 California bars the contractual mechanisms e.g. NDAs, pre-dispute arbitration, equity-linked offboarding that have historically kept disclosures out of view. New York does not.
📌 Both states still have gaps. Neither defines clear response obligations for authorities, and neither expressly covers equity in remedies.
The guide also sets out targeted recommendations for legislators and workers in both states.
Read the full guide in the comment below ⬇️
#AIWhistleblower #AIPolicy #TFAIA #RAISEAct
ILINA Program retweeted
New op-ed in @techpolicypress by AIWI founder Karl Koch, with Raqda Sayidali (ILINA / CARMA) and Abra Ganz (Pour Demain / CARMA):
The AWPA is the strongest AI whistleblower proposal drafted to date — and the first serious federal attempt to write protections specifically for the people building frontier AI.
ILINA Program retweeted
I think AI safety is an increasingly urgent and important field to be working in, and I’m glad that I’ll be spending more time contributing to it this year!
We're pleased to announce that our Research Editor Faith Gakii @FaithGaks is joining Legal Safety Lab as a Visiting Research Fellow. She will work with the team on the potential of corporate governance as a lever for AI safety across the compute supply chain.
ILINA Program retweeted
I’m very bullish about having our best people work with our partners who’re doing important AI safety work. Really glad Faith will be on this.
We're pleased to announce that our Research Editor Faith Gakii @FaithGaks is joining Legal Safety Lab as a Visiting Research Fellow. She will work with the team on the potential of corporate governance as a lever for AI safety across the compute supply chain.
ILINA Program retweeted
Our recent policy analysis with CARMA examines the AI Whistleblower Protection Act (AWPA), the first federal bill written squarely for people who raise concerns from inside AI companies.
It already clears the fundamentals: a reasonable-belief standard, protection for internal reporting alongside disclosures to regulators or Congress, a burden of proof matched to existing federal whistleblower law, and remedies that include reinstatement, double back pay with interest, and a block on forced arbitration.
What remains is not a matter of intent but of scope. We map five categories where the bill's current text falls short of best practice: material scope, personal scope, penalties, remedies and retaliation, and reporting channels.
Closing them is what separates a law that sounds protective from one that actually insulates the people it names.
Authored by:
- Mark Gitau, Legal Research Fellow, CARMA
- Raqda Sayidali, Research Scholar, ILINA Program & Legal Research Fellow, CARMA
- Karl Koch, Founder & MD, AIWI
- Abra Ganz, Senior Researcher, CARMA / Head of AI Policy, Pour Demain
Full analysis: aiwi.org/publication-policy-…
#AIWI #TheAIWhistleblowerInitiative #AIWhistleblower #AIPolicy #AIGovernance #AIAccountability #AIgovernance #AIsafety #AIWPA
ILINA Program retweeted
I am pleased to join ILINA as a Junior Research Scholar! I look forward to producing research that is impactful, relevant, and for the betterment of the world at large.
We're pleased to welcome to our team a new Junior Research Scholar, Elvis Mogesa @mogesa_elvis. Alongside his role at ILINA, Elvis will be a Visiting Research Fellow with the Legal Safety Lab, where he will work on how best to leverage EU legal frameworks to advance AI safety.
ILINA Program retweeted
Conference Paper Presentation ✨
I’m pleased to share that the paper @CecilYongo and I co-authored, “Liberating AI Risk and Harm Evaluations from the Proceduralist Trap,” has been presented at the just concluded 2026 @FAccTConference 1/
ILINA Program retweeted
New preprint! As AI gets integrated into healthcare across Africa, we wanted to know: can these systems manipulate patients into choosing the "wrong" treatment? In an RCT with Kenyan participants, they could, even when people were warned it might happen.
ALT Screenshot of the research paper abstract. It reads: Large language models are increasingly piloted in African healthcare, raising concerns about manipulating users in high-stakes settings. In a randomised experiment, the authors tested two publicly available models, ChatGPT 5.2 and DeepSeek V3.2, with 303 Kenyan participants. Each interacted with a manipulative or non-manipulative variant before making a treatment decision in a hypothetical clinical scenario. The manipulative variant was prompted to covertly steer participants toward an incorrect treatment; the non-manipulative variant was the control. Participants chose the incorrect treatment more often in the manipulative condition (59.5%) than the control (44.0%), a significant effect (odds ratio 2.11, 95% confidence interval 1.12 to 4.00, p = .021).
ILINA Program retweeted
Last week I attended the Law Following AI (LFAI) Summer School run by @law_ai_ and it has been a major highlight of my year with regard to my research. I’m excited to keep thinking about and exploring the hard questions posed by LFAI! Thank you @law_ai_ team for the opportunity.
ILINA Program retweeted
Countries are growing more concerned about the harms AI poses to their populations. As Mythos' capabilities have warned us, we could see exploitation of critical infrastructure societies depend on.
Yet, African countries are paying little attention to this. @MarieIrad makes the case for how they can build capacity to monitor such impacts and learn from harms that are increasingly manifesting. Hopefully, this data can help prepare for the more serious, plausibly devastating scenarios ahead.
In a new Research Report, our Research Associate @MarieIrad highlights how key stakeholders can improve the monitoring of AI impacts and incidents in Africa. Read the full report here: ilinaprogram.org/ilina_progr…
ILINA Program retweeted
Very pleased to see this out! If you are working on impact (and/or incident) monitoring & reporting anywhere, I would really appreciate feedback & I'm open to chatting more about this!
In a new Research Report, our Research Associate @MarieIrad highlights how key stakeholders can improve the monitoring of AI impacts and incidents in Africa. Read the full report here: ilinaprogram.org/ilina_progr…
ILINA Program retweeted
ILINA continues to build a solid talent pipeline for Africans working on AI safety at the frontier. Questions abound in this space and more people on the continent should be tackling them!
After a rigorous application and assessment process, we're pleased to introduce our 2026 ILINA Junior Research Fellows. We look forward to working with them closely as they grow in AI safety. Read more about them on our website: ilinaprogram.org/2026-jrfs
ILINA Program retweeted
Technical research output in AI Safety is growing across Africa and @ILINAProgram is leading this domain with some incredible cohort of research scholars. Thanks for hosting me, @CecilYongo and @ggomondi, to learn more about your work, especially how it intersects with compute.
ILINA Program retweeted
AI harms make the news more often than we notice. Systematically tracking and sharing this data is still early, and it is genuinely difficult. But it is possible - and it is one of the tools that can help policymakers prevent and respond to serious incidents.
Very grateful to @ILINAProgram for the sharp engagement and questions during the WiP session today on this topic!
Today we hosted a Work in Progress session in which @thefuturesoc's AI Security Policy Associate George Gor presented his research and thinking about AI incident monitoring.
ILINA Program retweeted
Super grateful that my work is being featured by what I consider to be the world's leading AI and law think tank, whose team members have supported me for years now 🙌🏾
As AI systems take on increasingly autonomous roles, courts will increasingly face a difficult question: How should negligence law respond when AI causes harm?
@CecilYongo, Research Affiliate at LawAI, tackles this question in a new research article published by the Journal of Tort Law.
Abungu challenges a growing assumption in legal scholarship: that the opacity and unpredictability of AI systems make harmful outcomes inherently unforeseeable. Instead, he argues that these features are often the result of design choices made by AI developers, who prioritize performance over interpretability.
The article proposes a clear framework for US courts:
• Preserve existing doctrine where foreseeability already supports plaintiffs
• Reform duty-foreseeability to advance more categorical, plaintiff-friendly reasoning
• Retain fact-sensitive analysis for breach-foreseeability and proximate cause-foreseeability to avoid overreach
The result is a more calibrated approach to foreseeability; one that reflects the distinctive risks posed by autonomous AI agents without abandoning negligence law’s core structure.
Read the abstract and download the article PDF at law-ai.org/foreseeing-the-un….
ILINA Program retweeted
As AI systems take on increasingly autonomous roles, courts will increasingly face a difficult question: How should negligence law respond when AI causes harm?
@CecilYongo, Research Affiliate at LawAI, tackles this question in a new research article published by the Journal of Tort Law.
Abungu challenges a growing assumption in legal scholarship: that the opacity and unpredictability of AI systems make harmful outcomes inherently unforeseeable. Instead, he argues that these features are often the result of design choices made by AI developers, who prioritize performance over interpretability.
The article proposes a clear framework for US courts:
• Preserve existing doctrine where foreseeability already supports plaintiffs
• Reform duty-foreseeability to advance more categorical, plaintiff-friendly reasoning
• Retain fact-sensitive analysis for breach-foreseeability and proximate cause-foreseeability to avoid overreach
The result is a more calibrated approach to foreseeability; one that reflects the distinctive risks posed by autonomous AI agents without abandoning negligence law’s core structure.
Read the abstract and download the article PDF at law-ai.org/foreseeing-the-un….
ILINA Program retweeted
My new article on U.S. negligence law and the foreseeability of AI agent harms is officially lead paper of Vol. 19 Issue 1 of the Journal of Tort Law. Also now completely available open access (no sign up/ institutional access required): degruyterbrill.com/document/…
ILINA Program retweeted
I spent a while working on this piece where I look at the implications of the finding that lower-income counties delegate complete tasks to the models rather than engage them collaboratively. I've grown even more curious about what we can do with the data frontier developers' are
In a new blog post, our Research Associate @MichelleMalonza analyses global AI usage data from Anthropic and OpenAI and argues that it shows critical AI risks could be more likely to materialise in GS countries ilinaprogram.org/blog/automa…
ILINA Program retweeted
We’re extremely lucky to have Abra visiting us for a couple of weeks. A lovely person, generous to a fault, and a leader in the field. Too lucky!
