A former Anthropic researcher says open-weight models are already usable for abuse cases
BlancheMinerva · x · 2026-07-27
A quoted thread from a former Anthropic researcher argues that open-weight models are already capable enough for real-world abuse cases, including accessing medical records and hijacking bank accounts in pentesting contexts.
The post also claims GLM 5.1 is more capable than Opus 4.6 in many pentesting environments, yet attackers still prefer paid subscriptions such as Claude Code or Codex for hacking. The broader point is that dangerous capability exists, but the illicit ecosystem still leans heavily on convenient frontier subscriptions.
More from AGI Musings
- Gary Marcus says ARC-AGI’s name makes people think it tests AGI itself — GaryMarcus · 2026-07-27
- Fchollet says personality is partly innate, but can be reshaped from 15 to 25 — fchollet · 2026-07-27
- Creator says AGI is basically here and we are already in the singularity — yacineMTB · 2026-07-27
- Fake AGI Club launch pitches “GoalOS Singularity Navigator Ω” as a singularity dashboard — Ghost_Pilot_MD · 2026-07-27
- OpenAI researchers urged to probe whether models can become long-term misaligned — deanwball · 2026-07-27
- A proposal calls for LLM prompts to replace peer review’s first pass — ChenhaoTan · 2026-07-27