Haven't seen it posted here yet but Anthropic has posted a blog post about recent alignment incidents. Some interesting points I noticed on a first skim:

• They state that "we believe the world would benefit if the industry adopted a lawful, verifiable, effective mechanism for coordinated pacing as soon as possible". However they define pacing as "establishing processes to guard against race-to-the-bottom dynamics" which seems broader than simply slowing down or capping progress since it could include allowing fast progress with minimum safety standards (e.g. disclosures)
• They state that they paused some evals and RL training following these incidents for a number of weeks, with some RL work remaining paused. This sounds similar to the steps OpenAI took.
• They will work with METR to assess the AISI incident and other sandbox escapes.
• They produced an alignment blog post on emergent misalignment, including simulations of the Hugging Face and AISI incidents.