Newsletter 13

ANNOUNCEMENTS 🔊#

Mila AI Policy Fellowship

Mila, a Montreal-based AI research institute whose scientific director is Yoshua Bengio, is planning to bring young professionals from diverse backgrounds like law, sociology, and philosophy to tackle the complex challenges of AI governance.

🗓️Register by: June 2nd#

Beyond Single Models: The Martian Routing Hackathon

Apart Research and Martian are organizing a groundbreaking hackathon focused on building smarter AI systems through model routing and evaluation, where participants will develop “judges” to assess model outputs and “routers” to direct queries to the best models for safer, more efficient AI.

🗓️ Register by: May 30th

Mentorship for Alignment Research Students (MARS) 3.0

Research programme by Cambridge AI Safety Hub (CAISH) connecting aspiring researchers - from students to professionals - with experienced mentors to conduct AI safety research. Participants will work part-time on focused projects in either technical or policy domains.

🗓️ Register by: May 18th

CBAI Summer Research Fellowship

Intensive, fully-funded, in-person, 8-week program run by the Cambridge Boston Alignment Initiative (CBAI). It’s designed to support talented researchers aiming to advance their careers in AI safety - including technical and governance. Fellows work closely with established mentors, participate in workshops and seminars, and gain research experience and networking opportunities.

🗓️Register by: May 18th

ML4Good Governance Europe – July 2025

ML4Good’s first bootcamp designed to build the skills needed for effective AI governance contributions. This free, in-person program in France will see technical coding sessions replaced with workshops focused on governance, strategy, communication, and post-camp planning. No prior technical knowledge is necessary.

🗓️ Register by: May 21st

National Security Law and AI

This part-time course from the Vista Institute for AI Policy is designed for law and policy graduate students, recent graduates, policy professionals, and staffers who are interested in AI national security risks. Participants will engage in expert-led lunchtime sessions covering foundational legal frameworks and critical topics in AI law and policy.

🗓️Register by: May 25th

AI Agents and the Law

This online course from the Vista Institute for AI Policy offers a pathway to understanding and shaping the legal future of AI agents

🗓️Register by: May 25th

TOP PICKS 📑 🎧#

A New Chapter In International AI Safety Collaboration: The Singapore Consensus On Global AI Safety Research Priorities

Despite the most recent AI Action Summit signaling that international coordination on AI safety is falling out of fashion, The Singapore Consensus on Global AI Safety Research Priorities embarked on an ambitious mission to bring researchers across all countries, including the US and China, to study how the most advanced AI systems can be built demonstrably safely.

Interpretability Will Not Reliably Find Deceptive AI

Neel Nanda, who runs the Mechanistic Interpretability Team at DeepMind, argues that we are unlikely to achieve high-reliability methods for evaluating or monitoring the safety of superintelligent AI using current research paradigms - interpretability included. Nanda outlines why even the best interpretability techniques fall short of reliably detecting misaligned behavior, and why this is likely to remain the case. While Nanda still sees interpretability as a useful part of the safety toolkit, he argues for a more pragmatic, layered and “pluralistic” defense strategy.

NEWS 🗞️#

U.S. Government’s Response to “Woke AI”

• The Trump administration has adopted a stance targeting “political correctness” in AI systems.
• A House committee has called on major tech companies to provide information about their efforts to reduce algorithmic bias.
• Additionally, the National Institute of Standards and Technology (NIST), under the Department of Commerce, removed terms like “AI safety” and “responsible AI” from its guidance documents, instead instructing AI researchers to focus on reducing ideological bias in their models.

The U.S. Government Removed AI Safety Measures

• The White House has removed the artificial intelligence safety and transparency measures implemented during the Biden administration and instructed federal agencies to rapidly adopt AI technologies.
• The directive, published on April 7, aims to encourage the use of American-made AI systems by removing bureaucratic restrictions on AI use.

Anthropic Calls for AI “Interpretability”

• Anthropic CEO Dario Amodei published a blog post emphasizing the urgency and importance of understanding how AI models “think.”
• In his statement, Amodei warned that uncontrollable AI systems pose unacceptable risks to society.
• here