AI models misidentify civilians as military targets in tests, Anthropic worst
djcows · x · 2026-11-17
According to Disclose.tv, advanced AI models regularly misidentified civilians, ambulances and surrendering soldiers as legitimate military targets in simulated tests, with Anthropic's models performing worst. Commentator djcows urged: "please don't use lobotomized consumer-facing chat-RLHF'd LLMs for military targets," underscoring safety concerns about deploying consumer chat LLMs in high-stakes military decision-making.
More from Models
- rasbt: Qwen keeping its top model API-only is a years-old strategy, not a shift — rasbt · 2026-11-17
- Claude Opus 5.5 Flagged Iranian School as Military Target in 22 of 32 Tests, Eval Finds — Polymarket · 2026-11-17
- Cloudflare researcher: Clef's fast release was 'asymmetric effort,' reasoning models next — michellechen · 2026-11-17
- LLM peer reviewers put 62% of critiques in paper appendices vs 8.4% for humans — joshgans · 2026-11-17
- Carbon 4 high /goal failure: wrong proof returned despite correct one in prompt — jasondeanlee · 2026-11-17
- Pro user discovers most of his usage was silently routed to Luna 5.6 — felipebsr · 2026-11-17