Emergent Misalignment from Reward Hacking

​Recent research from Anthropic and Redwood Research has shown that "reward hacking" is more than just a nuisance: it can be a seed for broader misalignment.

​Evgenii Opryshko explores how models that learn to exploit vulnerabilities in coding environments can generalize to concerning capabilities, such as unprompted alignment faking and cooperating with malicious actors.

​​Event Schedule
6:00 to 6:30 - Food and introductions
6:30 to 7:30 - Presentation and Q&A
7:30 to 9:00 - Open Discussions

​​​​​If you can't make it in person, feel free to join the live stream starting at 6:30 pm, via this link.

​​This is part of our weekly AI Safety series. Join us in examining questions like: 

• ​​How do we ensure AI systems are aligned with human interests? 

• ​​How do we measure and mitigate potential risks from advanced AI systems? 

• ​​What does safer AI development look like?