GAO audit finds Claude, Grok, ChatGPT and Gemini safeguards easily bypassed
midwoodgirl10 · reddit · 2026-11-19
A Reddit post cites an NBC News report on a Government Accountability Office audit: testing the most-used generative AI platforms — Claude, Grok, ChatGPT and Gemini — found safeguards easily bypassed, with users able to obtain detailed attack plans within 12 hours, including targets, locations, timing and methods.
The poster asks: if safeguards can be so easily bypassed, is alignment even achievable, and shouldn't it be solved before newer models ship? Discussion follows on the gap between alignment ideals and real-world guardrails.
More from Models
- Abacus AI open-sources uncensored Smaug, a 27B model aimed at agentic work — bindureddy · 2026-11-19
- Harmonic's Aristotle rewrites Bitcoin Core's C signature parser in Rust with a formal proof — satnam6502 · 2026-11-19
- GPT-6.2 Astra users note it can work 30 minutes without any progress updates — Angaisb_ · 2026-11-19
- Arena points readers to the live Agent Arena Index and its methodology — arena · 2026-11-19
- LMArena launches six new agent indexes; Claude Fable 5.2 tops Work, Coding and Chat — arena · 2026-11-19
- Investor complains of 'rude' Anthropic rate limit, blames Nvidia deal monopoly — StewartalsopIII · 2026-11-19