Events

Applications for MAIA’s Fall 2026 AI Safety Fundamentals fellowship are due Wednesday, September 23 at 11:59 PM Eastern Time. Apply: airtable.com The eight-week reading and discussion fellowship begins the week of September 28. Cohorts meet weekly in person at MAIA’s office in Kendall Square. Each two-hour session includes reading time, so no additional work is required outside of sessions. Program information: aialignment.mit.edu

MIT Stratton Student Center, 84 Massachusetts Ave, Cambridge, MA 02139, USA

MIT AI Alignment (MAIA) is bringing a Bon Me food truck in front of the student center on Tuesday (9/15) and Wednesday (9/16), 11 AM to 2 PM both days. Free food!! First come, first served. It’s more important than ever to ask questions about what AI will do in the next few years. MAIA members will be tabling near the food truck, many of whom have done safety research or worked in safety roles at AI labs and other organizations. Come talk to us about things like: • What do current trends in AI progress look like? • What do they mean for the next few years? • Why are so many AI experts worried? • What are some of the ways to address these risks? Curious, or just hungry? We hope to see you there.

MIT Stratton Student Center, 84 Massachusetts Ave, Cambridge, MA 02139, USA

MIT AI Alignment (MAIA) is bringing a Bon Me food truck in front of the student center on Tuesday (9/15) and Wednesday (9/16), 11 AM to 2 PM both days. Free food!! First come, first served. It’s more important than ever to ask questions about what AI will do in the next few years. MAIA members will be tabling near the food truck, many of whom have done safety research or worked in safety roles at AI labs and other organizations. Come talk to us about things like: • What do current trends in AI progress look like? • What do they mean for the next few years? • Why are so many AI experts worried? • What are some of the ways to address these risks? Curious, or just hungry? We hope to see you there.

Essex Woods Retreat Center, Essex, Massachusetts

A summer residential AI safety workshop hosted by AISST and MAIA at Essex Woods. Students joined researchers including Nikola Jurkovic, Stephen Casper, David Bau, Tzu Kit Chan, James Sanders, Alex Turner, and Zac Hatfield-Dodds for talks, small-group discussions, one-on-ones, and conversations about AI safety research and careers. Workshop program: aisst.ai

Kendall Square Urban Park Roof Garden, 325 Main Street, Cambridge

MAIA’s end-of-year gathering brought the community together at the Kendall Square rooftop park. The social featured Indian food and space for conversation, icebreakers, and board games, with vegetarian, vegan, and gluten-free food options. It was open to MIT community members and affiliates. Original Partiful event: Partiful event

CIC, Cambridge

A one-day project sprint in AI control, interpretability, and strategy, with team presentations and prizes.

Cambridge, Massachusetts

MAIA and MIT’s Science Policy Initiative brought their communities together for an evening of pizza, games, and conversation. The joint social helped members meet students working on science policy and discuss shared interests in AI policy. Original Partiful event: Partiful event

MIT, Room 35-225

Andreea Bobu, an MIT AeroAstro and CSAIL professor who leads the CLEAR Lab, presented a framework for AI systems that understand the concepts people care about. Four Pillars of Human-Aligned AI Representations focused on completeness, minimality, efficiency, and interpretability. Drawing on robot learning, the talk examined generating concepts as code with language models and using language-guided abstractions to ignore irrelevant information, connecting these methods to meaningful human oversight. Original Luma event: Luma event Archival date-only entry: the complete start/end times are unknown. The source records a start time of 17:00 Eastern; the exact end time is unknown.

Essex Woods Retreat Center, Essex, Massachusetts

A joint AISST/MAIA residential AI policy workshop at Essex Woods. Talks and panels with AI policy practitioners included Nathan Calvin, Jason Hausenloy, Gabriel Weil, and Fin Moorhouse. The program also featured Harvard Law School and RAND panels, a tabletop exercise or wargame, career discussions, small groups, and one-on-ones. Workshop program: aisst.ai

CIC, Gooseberry, Cambridge

MAIA's member meeting with Max Tegmark combined dinner with discussion about AI safety. Tegmark, an MIT physics professor and co-founder of the Future of Life Institute, brought a research perspective spanning mechanistic interpretability and provably safe systems. This was a smaller member gathering, distinct from his public February speaker event. Original Partiful event: Partiful event

CIC, Floor 11, Gooseberry, Cambridge

Phillip Christoffersen, a PhD researcher in MIT CSAIL's Algorithmic Alignment Group, presented Beyond Guardrails: Towards a Common Specification Language for AI Agent Domains. The talk examined why existing guardrails often constrain outcomes rather than processes and introduced a specification language for describing agent environments and safety requirements. It explored generating training data for safety-constrained policies and using coding agents to make specification more scalable, with implications for oversight of language-model agents. Original Partiful event: Partiful event

Essex Woods Retreat Center, Essex, Massachusetts

The second spring AISST/MAIA technical AI safety workshop brought students and researchers together at Essex Woods. The weekend featured talks by David Bau, Abra Ganz, Joshua Clymer, Aryan Bhatt, Alex Turner, Nikola Jurkovic, and Dillon Plunkett, alongside small-group discussions, paired one-on-ones, and sessions on research and career next steps. Workshop program: aisst.ai

CIC, Floor 11, Cambridge; Josh Clymer joining virtually

Josh Clymer of OpenAI joined MAIA remotely for a member meeting on recursive alignment: keeping AI systems aligned through successive rounds of modification and self-improvement. The session drew on his technical safety work, including earlier research at Redwood Research and METR on AI control, alignment faking, and evaluations of self-improvement capabilities. Original Partiful event: Partiful event Archival date-only entry: the complete start/end times are unknown. The source records a start time of 17:00 Eastern; the exact end time is unknown.

CIC, Floor 11, Cambridge

Aryan Bhatt of Redwood Research joined MAIA for a member meeting on AI control: how to use powerful AI systems safely even when they might pursue hidden goals. The discussion included Redwood's proposal for making deals with early scheming systems, shared during the meeting. Bhatt's work spans control evaluations, scheming, and model organisms of misalignment. Related research article: blog.redwoodresearch.org Archival date-only entry: the complete start/end times are unknown. The source records a start time of 17:00 Eastern; the exact end time is unknown.

Essex Woods Retreat Center, Essex, Massachusetts

A joint AISST/MAIA residential technical AI safety workshop at Essex Woods. Sessions examined AI control, propensity evaluations, interpretability, human intent in software, and challenges posed by brain-like AGI. The weekend combined researcher talks with small-group discussions, one-on-ones, career planning, and informal time together. Speakers / session contributors: Sydney Von Arx, David Bau, Tzu Kit Chan, Kaivu Hariharan, Aryan Bhatt, Steven Byrnes. Workshop program: aisst.ai

CIC, 3rd floor, Cambridge

MAIA members and AI Safety Fundamentals fellows celebrated the fellowship’s first week with pizza, drinks, and snacks. The social offered a chance to get to know cohort members beyond the sessions and continue conversations about the readings and AI alignment. Original Partiful event: Partiful event

CIC, 12th floor, Cambridge

MAIA welcomed the newest AI Safety Fundamentals cohort with a social at CIC. Fellows and existing community members met over pizza, drinks, and snacks, with time for casual conversation about AI safety and the coming fellowship. Original Partiful event: Partiful event Historical timing note: this three-hour time window is an organizer-provided estimate.

MIT, Room 32-141, with Daniel Kokotajlo joining virtually; online option

Daniel Kokotajlo joined MAIA for a virtual Q&A on AI forecasting, the pace of AI development, and risks from misaligned systems. A former OpenAI researcher and co-author of AI 2027, he discussed questions about when transformative AI might arrive and how to reason about what comes next. The audience joined from MIT's 32-141 and online. Original Luma event: Luma event

CIC, 12th floor, Cambridge

MAIA and research fellows from the Cambridge Boston Alignment Initiative met for a joint social at CIC. Pizza, non-alcoholic drinks, and informal conversation gave the two AI safety communities time to get to know one another outside their regular programs. Original Partiful event: Partiful event Historical timing note: this three-hour time window is an organizer-provided estimate.

The first session of a planned ten-week technical AI safety upskilling program. Later sessions are not reconstructed without their individual records.

MIT Memorial Lobby (Lobby 10), Booth 1

Meet MIT AI Alignment in Lobby 10 and learn about the club’s AI safety activities and ways to get involved. Listed hours reflect the confirmed table reservation.

MIT, Room 26-100; Nick Bostrom virtual, Max Tegmark in person; online option

Nick Bostrom, author of Superintelligence and founding director of Oxford's Future of Humanity Institute, joined Max Tegmark for a fireside conversation about advanced AI and humanity's future. Bostrom joined remotely while Tegmark was on stage at MIT. The event connected longstanding arguments about existential risk with the questions raised by rapidly improving AI, followed by audience questions. Original Luma event: Luma event

MIT, Room 10-250, with speaker joining virtually and an online audience

Nicholas Carlini of Anthropic joined MAIA virtually for an AI security talk and audience Q&A. The event introduced his work on vulnerabilities in AI systems, including adversarial examples, language-model jailbreaks, and failure modes that make models difficult to secure. Students gathered in MIT's 10-250 to watch together, with an online attendance option. Original Luma event: Luma event Archival date-only entry: the complete start/end times are unknown. The source records a start time of 18:00 Eastern; the exact end time is unknown.

MIT Memorial Lobby (Lobby 10), Booth 1

MAIA tabling in MIT’s Lobby 10 to promote upcoming speaker events and introduce students to MIT AI Alignment. Listed hours reflect the confirmed table reservation.

MIT, Room 32-144

MAIA opened its spring member-meeting series with Stephen Casper on what makes AI safety research impactful. The talk was followed by a pizza social, giving members a chance to meet one another and discuss research directions before the semester's workshops, technical upskilling, and speaker events. Original Partiful event: Partiful event Scheduled talk: 5–6 PM. The pizza social followed afterward.

CIC, 12th floor, 245 Main Street, Cambridge

MAIA’s final social of the fall semester brought pizza, non-alcoholic drinks, board games, and conversation to CIC. The gathering was intended as a chance to reconnect with the community and relax before finals. Original Partiful event: Partiful event

MIT, Room 3-370

MAIA members and AISF fellows competed to devise questions that people could solve but leading language models would get wrong. Mission Strawberry put human teams up against GPT-5-Thinking and Gemini 3 in a playful test of model reasoning, with food and strawberry-themed treats. The organizer's post-event recap highlighted four submissions that fooled both models while every human team found the correct answer. Original Partiful event: Partiful event Mission Strawberry activity: mission-strawberry.replit.app Archival date-only entry: the complete start/end times are unknown. The source records a start time of 18:00 Eastern; the exact end time is unknown.

MAIA office, CIC, Cambridge

AISF fellows spent a full day turning ideas from the fellowship into hands-on AI safety projects, with team formation, mentorship, meals, and evening presentations. Documented projects included adding a small learned module to a language model to resist jailbreaks, using Petri to compare lying and reward hacking under ordinary versus competitive task settings, and building an agent to demonstrate manipulative persuasion. Teams presented their work at the end of the day and participants voted on projects. Speakers / session contributors: Riya Tyagi, Ionel Chiosa.

Essex Woods Retreat Center, Essex, Massachusetts

The second fall AISST/MAIA residential workshop at Essex Woods combined research talks with small-group discussions and paired one-on-ones. Participants explored technical AI safety, including mechanistic interpretability and AI control, discussed research and career directions, and connected with students from other university safety communities. Speakers / session contributors: Sam Marks, David Bau, Tyler Tracy, Aryan Bhatt. Workshop program: aisst.ai

Essex Woods Retreat Center, Essex, Massachusetts

The first fall AISST/MAIA residential workshop at Essex Woods brought students together with AI safety researchers for talks, small-group discussions, and one-on-ones. Sessions explored scheming and AI systems that support human understanding, alongside career reflection, planning, and informal conversations. Speakers / session contributors: David Bau, Aryan Bhatt. Workshop program: aisst.ai

CIC, 12th floor, 245 Main Street, Cambridge

A Friday-evening MAIA pizza social at CIC, open to community members and their friends. The event focused on informal conversation and meeting others interested in AI safety at the start of the weekend. Original Partiful event: Partiful event Historical timing note: this three-hour time window is an organizer-provided estimate.

MIT, Room 3-333

A MAIA movie night featuring Dr. Strangelove, the Cold War satire about nuclear escalation and decision-making. The screening offered a social evening around a film concerned with technology, institutions, and catastrophic risk.

MIT, Room 1-150

A joint MAIA and Impact@MIT social welcoming the wider MIT community. Over pizza, the gathering connected students interested in AI safety with students exploring positive impact through careers, donations, and other projects. Membership in either club was not required. Original Partiful event: Partiful event

https://maps.app.goo.gl/vNAx1jgKQ3bzbBGC7

View this event on Partiful at Partiful event

CIC, 12th floor, Cambridge

An end-of-year MAIA social centered on saying farewell to graduating members and sharing plans for the summer and life after MIT. The gathering offered pizza and a break from the lead-up to finals. Original Partiful event: Partiful event

Essex Woods Retreat Center, Essex, Massachusetts

A joint AISST/MAIA residential technical AI safety workshop at Essex Woods, featuring talks and an interview with researchers from Anthropic, Redwood Research, METR, Northeastern University, and the University of Texas at Austin. Small-group discussions, paired one-on-ones, and career sessions gave students opportunities to explore research questions and next steps. Speakers / session contributors: Sam Marks, Brad Knox, David Bau, Cody Rushing, Adam Kaufman, Kevin Liu, Neev Parikh. Workshop program: aisst.ai

Essex Woods Retreat Center, Essex, Massachusetts

A residential AISST/MAIA AI policy workshop at Essex Woods. Talks and panels examined US-China AI competition, AI rights and human safety, short timelines, policy careers, and AI strategy. Participants also took part in an AGI wargame, small-group discussions, and one-on-ones with researchers and policy practitioners. Speakers / session contributors: Peter Salib, Sandhini Agarwal, Nitarshan Rajkumar, Joel Burke, Thomas Larsen, Arushi Gupta, Kevin Wei, Mackenzie Arnold, Chris Byrd, Corin Katzke. Workshop program: aisst.ai

Essex Woods Retreat Center, Essex, Massachusetts

A joint AISST/MAIA technical AI safety workshop at Essex Woods. Researchers and students discussed large-scale AI, model evaluations, timelines, superintelligence and international security, and strategies for safely relying on increasingly capable AI. The residential weekend included talks, small groups, one-on-ones, and career planning. Speakers / session contributors: David Bau, Martin Wattenberg, Corin Katzke, Zac Hatfield-Dodds, Nikola Jurkovic, Joshua Clymer. Workshop program: aisst.ai

Ajeya Cotra of Open Philanthropy joined the AI safety community for a Q&A on rapidly improving AI and its effects on the world. The event drew on her forecasting work and experience leading Open Philanthropy's technical AI safety program, offering participants a chance to ask questions about AI progress and how to make its impact positive. Archival date-only entry: the complete start/end times are unknown. The source records a start time of 18:00 Eastern; the exact end time is unknown.

CIC, 12th floor, Cambridge

MAIA and Harvard’s AI Safety Student Team met for a joint evening social at CIC. The gathering brought the two student AI safety communities together for food and conversation, following MAIA’s member meeting with Leo Gao. Historical timing note: this three-hour time window is an organizer-provided estimate.

CIC, Floor 11, Cambridge

MAIA members gathered for a Q&A with Leo Gao before the joint MAIA and AISST social. The member meeting began on CIC's eleventh floor and was scheduled to continue into the evening social.

Hogsmeade, second floor

Technical AI safety coworking with the ARENA curriculum and TA support.

CIC, 12th floor, 255 Main Street, Cambridge

A MAIA community social with pizza, drinks, and board games at CIC. The invitation welcomed both returning and new members for a relaxed evening of conversation and games. Historical timing note: this three-hour time window is an organizer-provided estimate.

CIC, 12th floor, 255 Main Street, Cambridge

A December MAIA social with food, desserts, and board games at CIC. The event invited members to catch up near the end of the semester and talk about winter plans. Historical timing note: this three-hour time window is an organizer-provided estimate.

CIC, Floor 11, Cambridge

Stephen Casper, a MAIA graduate-student member, spoke about pitfalls in evidence-based AI policy. The member talk invited discussion of how evidence is used to guide AI policy and where that approach can go wrong, with food provided for attendees. Archival date-only entry: the complete start/end times are unknown. The source records a start time of 18:00 Eastern; the exact end time is unknown.

Essex Woods Retreat Center, Essex, Massachusetts

The second fall AISST/MAIA residential workshop at Essex Woods brought technical and policy perspectives together. Talks covered interpretability, sparse autoencoders, AI R&D evaluations, compute governance, and the development of technical safety projects. Students also joined researcher discussions, one-on-ones, and an AGI planning exercise. Speakers / session contributors: Hidenori Tanaka, David Bau, Neev Parikh, Sam Marks, Kevin Liu, Leo Gao, Janet Egan, Nikola Jurkovic. Workshop program: aisst.ai

CIC, 12th floor, 255 Main Street, Cambridge

MAIA members gathered at CIC for food, refreshments, and board games. The social made space to catch up during the fall semester and discuss plans for the coming winter. Historical timing note: this three-hour time window is an organizer-provided estimate.

Essex Woods Retreat Center, Essex, Massachusetts

A residential AISST/MAIA workshop combining technical AI safety and governance perspectives. Sessions at Essex Woods covered AI control, evaluations of AI research and development, interpretability, resilience, and planning for advanced AI, with one-on-ones, small groups, and career discussions. Speakers / session contributors: Buck Shlegeris, Lucas Sato, Alan Chan, Richard Ngo, Lawrence Chan, David Bau, Nikola Jurkovic, Leo Gao. Workshop program: aisst.ai

AISST office, 66 Church Street, Cambridge

MAIA and AISST hosted an informal lunchtime talk with David Krueger, then an assistant professor at the University of Cambridge working on AI safety. Members had an opportunity to engage with a researcher whose work spans reward modeling and gaming, interpretability, backdoors, and model distillation. The talk was followed by AISST's AI forecasting contest at the same venue. Archival date-only entry: the complete start/end times are unknown. The source records a start time of 12:00 Eastern; the exact end time is unknown.

CIC, Floor 11, Gooseberry, Cambridge; speaker virtual

Stephen McAleer of OpenAI joined MAIA virtually to discuss language-model agents that can carry out general tasks on computers. His talk connected the promise of these agents, including their potential to accelerate AI research, with the open problem of deploying them safely. The announced discussion covered his agent research and directions for alignment and control, with members gathering at CIC over dinner. Archival date-only entry: the complete start/end times are unknown. The source records a start time of 18:00 Eastern; the exact end time is unknown.

Essex Woods Retreat Center, Essex, Massachusetts

A joint AISST/MAIA technical AI safety weekend at Essex Woods, with talks on AGI timelines, threat models, model evaluations, preparedness, and how technical safety projects develop. Researchers joined students for moderated Q&A, small-group discussions, career planning, and one-on-ones. Speakers / session contributors: Daniel Kokotajlo, Lucas Sato, Zac Hatfield-Dodds, Sam Marks, Tamay Besiroglu, Neil Chowdhury, Nikola Jurkovic, Romeo Dean. Workshop program: aisst.ai

MIT, Room 32-155

Essex Woods Retreat Center, Essex, Massachusetts

A joint AISST/MAIA residential policy workshop at Essex Woods. Sessions explored AI liability, compute governance, national security, AI timelines, and the relationship between technical safety research and policy. The weekend also included a regulatory-frameworks panel, policy career discussions, small groups, and paired one-on-ones. Speakers / session contributors: Trevor Levin, Gabriel Weil, Timothy Fist, Sam Marks, Stephen Casper, Daniel Kokotajlo, Chris Byrd. Workshop program: aisst.ai

Essex Woods Retreat Center, Essex, Massachusetts

A joint AISST/MAIA residential weekend at Essex Woods exploring technical AI safety, AI progress, and research careers. Talks covered AI control, adversarial robustness, responsible scaling policies, threat models, and forecasting, alongside small-group discussions, paired one-on-ones, and time to connect with researchers and other students. Speakers / session contributors: Thomas Larsen, David Bau, Ethan Perez, Daniel Kokotajlo, Buck Shlegeris, Adam Jermyn, Nikola Jurkovic, Romeo Dean. Workshop program: aisst.ai

MIT, Room 36-112

MIT, Room 36-112

MIT, Room 36-112

MIT, Room 36-112

MIT, Room 36-112

HAIST office, Cambridge, Massachusetts

Zac Hatfield-Dodds of Anthropic led a practical mini-workshop for MAIA and AISST members on contributing to open-source projects. Participants learned how to choose a project and issue, read contributor guides, set up a development workflow, test changes, and work with maintainers through pull requests. Examples included NumFocus projects such as NumPy, pandas, and Matplotlib.

Essex Woods Retreat Center, Essex, Massachusetts

A joint HAIST/MAIA residential AI safety workshop, bringing the Harvard and MIT communities together for a weekend of discussion and connection. Workshop program: aisst.ai

Essex Woods Retreat Center, Essex, Massachusetts

A joint HAIST/MAIA residential AI safety workshop, bringing the Harvard and MIT communities together for a weekend of discussion and connection. Workshop program: aisst.ai

6-120 (MIT Campus, Building 6, Room 120)

Max Tegmark is a professor doing AI and physics research at MIT as part of the Institute for Artificial Intelligence & Fundamental Interactions and the Center for Brains, Minds and Machines. He advocates for positive use of technology as president of the Future of Life Institute. He is the author of over 300 publications as well as the New York Times bestsellers “Life 3.0: Being Human in the Age of Artificial Intelligence” and "Our Mathematical Universe: My Quest for the Ultimate Nature of Reality". His most recent AI research focuses on mechanistic interpretability as well as news bias detection with machine-learning. Join with Google Meet: meet.google.com Learn more about Meet at: support.google.com

CIC Cambridge @ 245 Main, 245 Main St, Cambridge, MA 02142, USA

At the MAIA Office

MIT, Room 2-105

This Tuesday (9/5) at 5pm, there's Reward Hacking for Fun and Profit in room 2-105. Think like an RL agent and write stuff GPT-4 loves. The catch? You've no clue what GPT-4 wants. Winners get prizes, but there's boba and Chipotle for everyone!

W34-JIC (Johnson Ice Rink)

MIT, Room 2-131

This Monday (9/4) at 2pm, you can come to MAIA's AI Safety Demo Showcase in room 2-131. MAIA folks will be showing off AI safety projects, from "evil" versions of language models to adversarial policies that allow mere mortals to beat "superhuman" models at Go!

MIT, Room 3-133

This Saturday (9/2) at 3:30pm, we're doing GPT BattlePrompting in room 3-133. The challenge? Protect your virtual bank account while attempting to steal from others'. The twist? GPT-4 is guarding the passwords. If you're crafty, there's a cash prize up for grabs!

MAIA Office (245 Main St, Cambridge, MA 02142)

MAIA Office (245 Main St, Cambridge, MA 02142)

MAIA Office (245 Main St, Cambridge, MA 02142)

CIC Hogsmeade Floor 2

MIT 32-141

An AI safety talk with Richard Ngo of OpenAI.

MAIA Office (245 Main St, Cambridge, MA 02142)

MIT E25-117

A talk and Q&A with OpenAI researcher Daniel Kokotajlo. 📝 Details: • Date: Tuesday, March 21st • Time: 5–6 pm • Location: Room E25-117 🌟 Highlights: • Talk mentioning: • A brief history of AGI development efforts • How we got to GPT-4 • Safety implications of GPT-4 and similar AI systems • Q&A session with Daniel Kokotajlo • Free food

MAIA Office (245 Main St, Cambridge, MA 02142)

MAIA Office (245 Main St, Cambridge, MA 02142)

245 Main St, Cambridge, MA 02142, USA

We’re celebrating the official opening of the new MAIA office at 7pm this Saturday, Feb 11! Come for dinner (snacks and drinks also provided) and conversations with cool people. Feel free to bring other friends who are interested in MAIA! The office is at 245 Main Street, Cambridge; if you need access, call or text Alexandra at (339) 222-4211 or Shreya at (609) 558-5562 to be let in.