Abliterated DeepSeek DS4.1-1017 surfaces novel vulnerability missed by frontier models

cyb3rops · x · 2026-10-20

Security researcher cyb3rops ran VulnHunter with an abliterated DeepSeek DS4.1-1017 and reports one novel vulnerability that no frontier model found.

• Method: beyond benchmarks, models were pointed at a pinned, known-vulnerable production Go repo, with Grok as the frontier baseline.
• The results table shows 7 DS findings: one Unique (single-enforcement-point reviewer write guard, CWE-863, exploit-tested and correctly rated Low), the rest mostly overlapping Grok/GLM findings but with more conservative severity ratings.
• Trace files show DS examined every path — the gap isn't coverage, it judged the mechanism real but dismissed the impact.

Original post →

More from Models

• ChatGPT web can actually listen to audio, but the same model can't via API or Codex — Possible_Statement84 · 2026-10-23
• MiniMax M3.1 Flash free for 3 days on 0G, with 1M context and 512K output — MiniMax_AI · 2026-10-23
• TLA+ expert corrects Grok: TLC only exhausts finite state spaces, not all behaviors — tianyin_xu · 2026-10-23
• OpenAI blocks manual reset with 13% usage left: 'no reset needed right now' — AccomplishedSugar490 · 2026-10-23
• Meta's 816B Muse Spark Open scores 39, losing to 180B Qwen3.8-Flash-Next at 40 — BTA_Labs · 2026-10-23
• Gemini refuses to generate image in Docs, blames document size limit — viksit · 2026-10-23

Models channel →