Anthropic's Claude now self-assesses conversations for misuse, sparking surveillance concerns
sebkrier · x · 2026-10-31
- Anthropic announced "self-assessment": after each response, Claude produces a preliminary safety assessment of the conversation to flag potential misuse. In use since June, it has surfaced misuse that existing safeguards weren't designed to catch.
- Why it differs from classifiers: instead of targeting a discrete set of harms, it leverages Claude's own judgment, improving as models get better — and it's efficient since the model has already read the conversation. Details are on Anthropic's new Safeguards blog.
- Pushback: NeerajKA quote-shared the announcement, questioning whether "a small group of people with a particular set of strongly held moral convictions" should surveil and control what is becoming "our thought space."
More from Safety
- UK Biobank set to reopen to researchers after data breach — zakkohane · 2026-11-01
- Judge allows NAACP lawsuit over xAI data center pollution to proceed — SnoozeDoggyDog · 2026-11-01
- Viral claim: Grok Bot found Claude activated mic 312 times overnight on iPhone — RachelVT42 · 2026-11-01
- On alignment and SURA critics: both can be shallower than they claim — JacquesThibs · 2026-11-01
- Finland opens criminal investigation into Google's forest clearing for datacentre site — SnoozeDoggyDog · 2026-11-01
- Open-source CBS sandboxes local LLM agents at the OS level, default-denying risky tool calls — Luoriyy · 2026-11-01