Center for Human-Compatible AI

📣 Open Call for Posters! Submit your work to the poster session at the CHAI 2026 Workshop. Link below! ⏱️ Deadline: March 26, 2026 at 11:59p.m. PST. 🗓 June 4–7, 2026 at the Asilomar Conference Grounds in Pacific Grove, CA.

Read our call for posters here! workshop.humancompatible.ai

We're interested in both emerging questions and in less recent research, if relevant.

📣 Open Call for Posters! Submit your work to the poster session at the CHAI 2026 Workshop. Link below! ⏱️ Deadline: March 26, 2026 at 11:59p.m. PST. 🗓 June 4–7, 2026 at the Asilomar Conference Grounds in Pacific Grove, CA.

How to elicit truth from models that may be mistaken❌ or deceptive😈? In our @CHAI_Berkeley paper @iclr_conf, we reward each model by how much its answer helps predict the others'. With weak supervision from a 0.14B LM, it enables anti-deception training on a 8B LM and overwhelmingly outperforms LLM-as-a-Judge. This technique, peer prediction, is adapted from the mechanism design literature, where it's known to be incentive-compatible, i.e., incentivizes honesty. The intuition is that, predicting mistakes/lies when you know the correct solution is relatively easy, while the opposite is asymmetrically hard. We are able to further show that, with a large and diverse pool of models, peer prediction incentivizes honesty even when the supervisor doesn't know the models' prior beliefs and motivations.

What if an AI could learn to hide its thoughts? We show that LLMs can learn a general skill to evade activation monitors, with 0-shot transfer to unseen deception/harmfulness monitors from the literature. We call these "Neural Chameleons". A thread on our new paper. 🦎🧵

Our NeurIPS 2025 paper extends adversarial learning (adversarial examples, self-play, etc.) beyond zero-sum games by solving "self-sabotage". 🧵👇

And, watch the AI Red Lines #CallToAction: youtu.be/_SZe4ci9Ycs

Learn about the Global Call: red-lines.ai

Today, @securite_ia, CHAI, and @thefuturesoc are joined by 70+ leading orgs & 200+ signatories in a global call for AI Red Lines. Together, we are calling for international agreement to prevent the most severe risks to humanity and global stability. #AIRedLines Learn more:

The Global Call for AI Red Lines is live!! More than 200+ former heads of state, Nobel laureates, and other respected thinkers and leaders, and 70+ organizations are together calling for “do not cross” limits re: AI’s most severe #risks

@Michael05156007 aprecruit.berkeley.edu/JPF05027 Learn more and apply here 🔗

We’re hiring a research assistant for the book that @Michael05156007 is writing on extinction risk from AI! Please apply by September 19, 2025. Link in the next tweet:

Our mentors work on a broad range of topics. Check them out here: humancompatible.ai/chai-internshi…

Our 2026 internship applications are now open! Learn more about the internship and apply: #chai-internship" target="_blank" rel="nofollow noopener">humancompatible.ai/jobs#chai-inte… Deadline: October 5, 2025, at 11:59 p.m. PST

Who should apply? Current undergrads, Master’s, and PhD students and researchers, researchers in CS or adjacent fields, professional software or ML engineers.... The list goes on! If you're highly motivated to make progress on AI safety, consider applying.

Our interns: • Contribute to research with the potential for paper authorship • Build a pathway into AI safety work • Work alongside curious and ethically minded researchers

*New AI Alignment Paper* 🚨 Goal misgeneralization occurs when AI agents learn the wrong reward function, instead of the human's intended goal. 😇 We show that training with a minimax regret objective provably mitigates it, promoting safer and better-aligned RL policies!

We built an AI assistant that plays Minecraft with you. Start building a house—it figures out what you’re doing and jumps in to help. This assistant *wasn't* trained with RLHF. Instead, it's powered by *assistance games*, a better path forward for building AI assistants. 🧵

(1/7) New paper with @khanhxuannguyen and @thetututrain! Do LLM output probabilities actually relate to the probability of correctness? Or are they channeling this guy: ⬇️

🚨Our new #ICLR2025 paper presents a unified framework for intrinsic motivation and reward shaping: they signal the value of the RL agent’s state🤖=external state🌎+past experience🧠. Rewards based on potentials over the learning agent’s state provably avoid reward hacking!🧵