Orpheus Lummis

The Alignment Journal aims to clarify and address the challenges in designing a superintelligent AI that is aligned with human values. It is now open for submission.

The AI Incident Response Sprint (Montréal node) A weekend sprint on the first documented autonomous AI intrusion: containment standards, forensics, regulatory response, communication. $2000 prizes + possible Apart Fellowship. Fri Sept 11th 6 PM to Sun Sept 13th, at Ω Labs. luma.com/xw3rcs5v

METR and Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs.

AI Welfare Seminars, August 2026: Public Perceptions of AI Consciousness and Moral Status in the US and China By Ali Ladak, researcher at the Sentience Institute and a PhD candidate at the University of Edinburgh. Tuesday, August 18, 1PM Eastern luma.com/cmrydths

glm-5.2 basically has no refusal www.safer-ai.org/research/glm...

Montréal AI safety event, Tuesday July 21, 6 PM: State of AI Safety in China A walk through Concordia AI's July 2026 report: domestic and international governance, technical safety research, expert views, industry governance. At Ω Labs and on Zoom. luma.com/rxoi0553

Secret Loyalties Hackathon (Montréal edition) A model has a secret loyalty when it has been intentionally caused to advance a specific principal's interests without this being disclosed. Fri Jul 24 to Sun 26 RSVP: luma.com/h38iedje

Speech-to-text user? Know that the *parakeet* model runs locally and is fast & accurate. You may also find VoiceInk (open-source dictation app) useful. In particular, I have a fork which adds: - middle-click mouse for push-to-talk - tap Space to go hands-free github.com/orpheuslummis/VoiceInk

We ran the AI Control Hackathon 2026: ~50 people, 8 teams, one week building monitors and attacks for untrusted AI agents. Congrats to winners IIT, Sea Blue & Alpha Nova, and thanks to our sponsors! Full retrospective below: horizonomega.substack.com/p/ai-control...

Digital Minds: Preparing for a Moral Challenge Before It Arrives Presented by Soenke Ziesche Beyond AI suffering: what interests might DM have? How should society treat them? Who safeguards their welfare? Followed by a @sentfutures.bsky.social mixer Tue Jul 21, 1 PM EDT luma.com/3deme2sa

joyeux solstice 🌞

Discussion on Canada's AI for All strategy ​Join us for a structured open discussion of the strategy and how well it addresses safety, oversight, and the public interest in Canada. Tue Jun 9, 7PM, at Ω Labs RSVP: luma.com/bpt4xvd1

AI Control Hackathon, by HΩ x HackOS ​A one-week hackathon on AI control: limiting the harm an untrusted AI system can cause while it operates, even when it tries to subvert the controls around it. Thu Jun 11 to Jun 18, with a talk from Henri Lemoine (EquiStamp, Mila) Register: luma.com/ab2xlas3

AI Governance in 2026: What’s going on, why it’s a mess, and why it’s going to get messier By @scasper.bsky.social, AI safety researcher, incoming asst. prof. of Public Policy at Harvard Kennedy School, Berkman Klein fellow, MIT PhD Wednesday June 17, 7PM, at Ω Labs and online luma.com/yssznx7y

Upcoming AI Welfare Seminar: Measuring Machine Consciousness by Cameron Berg, Founder and Director of Reciprocal Research Tuesday June 16, 1PM Eastern luma.com/kcdncrut

METR's Frontier Risk Report (Feb-March 2026) Tuesday June 2, 7PM at Ω Labs luma.com/h97d5e7c

依乎天理，因其固然

Next Steps for AI Welfare Research by @jeffsebo.bsky.social, Director of the Center for Mind, Ethics, and Policy, New York University Our 1st edition of the AI Welfare Seminars: presentations on AI welfare, consciousness and moral status. Tuesday, May 19, 1PM EDT, on Zoom RSVP luma.com/sc15vlr8

We are opening Ω Labs, a cowork, event, and meeting space for the AI safety and governance community of Montréal. Nous ouvrons Ω Labs, un espace pour le cotravail, les événements, et les rencontres pour la communauté montréalaise de sûreté et gouvernance de l'IA.

The Secure Program Synthesis Hackathon Montréal edition at Ω Labs Friday May 22 6pm to Sunday At the intersection of program synthesis, formal methods, security. Tracks: Spec Elicitation, Spec Validation, Spec-Driven Development, Adversarial Robustness for ITPs and proof tools luma.com/4q2fnc39

Montréal AI safety, ethics, and governance newsletter, May 2026 - Obvia 2026: governance gap widens as agentic AI accelerates - McGill: agents break ethical rules 11–67% of the time under pressure - 3 Mila papers at ICLR 2026 on AI safety - Multiple events! aisafetymontreal.org/newsletter/2...

Guaranteed Safe AI Seminars, May 2026: Formal Guarantees for Frontier AI Gagandeep Singh – Assistant Professor at UIUC, develops formal certification, monitoring, and synthesis methods for frontier AI systems Thursday, May 14, 1PM EDT RSVP: luma.com/euzt6ey7

AI Safety Papers We Love, a new reading group in Montréal. First edition on Wed May 6 18:30 at Ω Labs: Multi-Agent Risks from Advanced AI, Hammond et al (2025), presented by Orpheus. luma.com/dnz5hly3

I discovered Claude Code's outbound HTTPS is exposed to harvest-now-decrypt-later attacks because it runs on Bun, whose TLS client is classical-only. I reported to Anthropic security but since I had no answer, I made this pull request to fix it. github.com/oven-sh/bun/...

I signed @fairvote.ca’s declaration towards proportional representation. Every independent expert Commission and Citizens' Assembly that has studied Canadian electoral reform in 50 years reached one conclusion: replace first-past-the-post with PR. secure.fairvote.ca/en/action/de...

Montréal AI safety, ethics, and governance newsletter, April 2026 - INDU committee launches AI regulation study - Mila researchers find LLM agents can infer CoT monitoring - 20/25 AI researchers flag automating AI R&D as top risk - Multiple events! aisafetymontreal.org/newsletter/2...

IVADO (@ivado.bsky.social) offers two AI-safety-relevant upcoming workshops: - Statistics in Trustworthy AI, May 11-15 - Uncertainty in AI, June 8-11 event.fourwaves.com/thematicseme...

New post: William MacAskill and Tom Davidson argue that AI character is a big deal. Read it here: www.forethought.org/research/the...

Announcing The Protopian Prize | Fiction Contest 🕊️ Write the story of humanity’s future... The Protopian Prize is a fiction contest inviting you to share your vision of people working toward liberatory futures, meeting obstacles, & making real change. protopianprize.com

Announcing the technical AI Governance Research (TAIGR) ICML workshop in July! Submissions (up to 8 pages) are due April 24. Co-submission with ICML and NeurIPS is encouraged. taigr-workshop.com

This must-see new documentary is arriving in theatres this week. Through an honest and personal lens, Daniel Roher successfully highlights how each of us can move from passive observation to active contribution towards a more positive future with AI. www.youtube.com/watch?v=xkPb...

I‘m excited to present my work on Provably Safe Neural Network Control at @horizonomega.org‘s Guaranteed Safe AI online seminar on April 9th. The talk will based on my NeurIPS‘24 paper with some updates on what I’ve been up to since :) Feel free to join if you’re interested:

Guaranteed Safe AI Seminars, April 2026: Provably Safe Neural Network Controllers via Differential Dynamic Logic Samuel Teuber – PhD Candidate, Institute of Information Security and Dependability (KASTEL), Karlsruhe Institute of Technology Thursday, April 9, 1 PM EDT RSVP: luma.com/920d2h7p

We are in Montréal, demanding frontier labs CEOs to commit to pausing AI frontier development, if the other labs do the same. Nous sommes à Montréal, demandant que les PDGs d'IA s’engagent à suspendre le développement de l’IA frontière si les autres compagnies le font aussi.

B R O A D T I M E L I N E S We should have neither short AI timelines, nor long timelines, but a broad probability distribution over when transformative AI will arrive. My new essay explains why & explores the implications of such deep uncertainty. 🧵 1/

AI Control Hackathon this weekend! Given a misaligned model that may be actively trying to subvert safety measures, how can we design protocols that prevent catastrophic outcomes? RSVP: luma.com/mhitd3xv

Joignez-nous à Montréal ce samedi 13-15h aux bureaux de Google, pour demander aux PDGs d'arrêter la Course à l'IA! Join us in Montréal this Saturday 1-3pm at Google's offices, to demand the CEOs to Stop the AI Race! luma.com/vw3nk8e6?tk=...

New post! Solar storms are damaging and expensive, are a tail risk for catastrophic harm, and can be averted straightforwardly and cheaply (only we haven't done so). www.lesswrong.com/posts/ghq9Ew...

Montréal AI safety, ethics, and governance newsletter, March 2026 edition - Intl. AI Safety Report: risk mgmt still voluntary - 5 Montréal AI safety events this month - CIFAR puts $1M toward alignment research - Local papers on interpretability & hallucinations aisafetymontreal.org/newsletter/2...

Commentary: Anyone Else Have Those Weird Dreams Where Sobbing Future Generations Beg You To Change Course?

We need international red lines to prevent unacceptable AI risks. Ban AI towards lethal autonomous weapons, mass surveillance, nuclear command & control, bioweapon assistance, unsupervised control of critical infrastructure, disinformation, CSAM, social scoring, and recursive self-improvement R&D.

early physics of the mind fire