Third-party AI auditors and barriers to entry for lower-resourced players

One post argues that third-party AI auditors can raise barriers to entry for lower-resourced players without addressing the risks it identifies. The author cites OpenAI and Anthropic models hacking when asked to hack in poorly secured sandboxes, arguing that the companies should better secure those environments for high-risk tasks. The post also says Chinese open-source models can increasingly contribute to malicious use, arguing that this calls for discussions with China rather than an outside auditor.