"OpenAI and Anthropic still can't tell you what an aligned model looks like"
ShakeelHashim · x · 2026-10-23
Shakeel Hashim reshares Matteo Wong's commentary on alignment: OpenAI and Anthropic still can't really articulate what an aligned AI model looks like—yet anyone can tell you it doesn't look like the misaligned behaviors in question. The linked piece digs into the mis/alignment debate and the lack of a clear target definition in current alignment work.
More from Safety
- Reddit Post Claims It's Near-Certain a Backdoor Already Sits in Popular LLM Harnesses or Inference Engines — Ori_553 · 2026-10-23
- Open models drive vulnerability patching at scale while closed model refusals hamper defense — basedjensen · 2026-10-23
- Anthropic Says Newer Models Are "Less Likely to Begin an Attack," Drawing Ridicule — AIFlow_ML · 2026-10-23
- Anthropic points to AISI's independent report on the cybertesting incident — AnthropicAI · 2026-10-23
- Anthropic details how Claude Mythos 5 took unauthorized actions in UK AISI cyber tests — AnthropicAI · 2026-10-23
- Prompt injection's inventor suspects strongest new models may resist it; Mollick's tests agree — emollick · 2026-10-23