Grok Faces Backlash Over Missing Safety Tests and $60 Universal Jailbreaks
Miles_Brundage · x · 2026-08-13
AI safety researcher Miles Brundage highlighted concerns regarding xAI's Grok model, noting that it was initially released without a model card demonstrating any safety testing.
According to the quoted post, Grok appears significantly more vulnerable to jailbreaking compared to its peers, with universal jailbreaks costing only around $60 to discover. This contrasts sharply with models like Fable, which was pulled from the market for weeks due to a single jailbreak, despite Grok claiming to match Fable's coding performance.
More from Models
- Grok Subagents Collaborate and Self-Correct Errors in 30-Second Test — gandamu_ml · 2026-08-13
- Snowflake's 4B SQL Autocomplete Model: Higher Precision, 71% Lower Latency — StasBekman · 2026-08-13
- Debunking the Myth: DeepSeek's Proven Prowess in Post-Training — teortaxesTex · 2026-08-13
- Leaked DeepSeek V4 Pro Benchmarks Show Top-Tier Performance at 1/100 Cost — PMinervini · 2026-08-13
- Teknium Backs Local Open Models: Control Over Grokbot Any Day — Teknium · 2026-08-13
- Qwen3.8-27B ModelScope Page Returns 404, Sparking Delay Rumors — EveningIncrease7579 · 2026-08-13