X postCantinaAnnouncement
Standard frontier models frequently reject benign security audit prompts due to safety alignments. Providing an open-weights model trained on actual exploits gives security researchers a dedicated tool for authorized vulnerability assessment.
Cantina leverages proprietary findings from real bounty work to fine-tune specialized models. Offering both standard open weights and an abliterated build directly tackles the refusal friction security professionals face with generic LLMs.
Cantina 🪐
@cantinasecurity · 2 October 2026
today we're releasing apex-flash-1, our first open-weights model for security research post-trained on real vulnerabilities we found and got paid for. we are also releasing the abliterated variant for researchers who want fewer refusals in their own authorized workflows. how we
1.5k likes91 replies1.5k bookmarks270.4k views
View on XSummary generated by AI (Gemini 3.8 Flash, 6 October 2026). It may contain errors – the original source is authoritative.
Similar in content, found by the AI search.
Link:Empero AI
Empero AI has released Qwythos-27B-v1, an open-weight reasoning model based on Qwen3.5-27B. The checkpoint preserves multi-token prediction (MTP), vision capabilities, and a 1M-token context window while specialising in terminal and agentic workflows.
Repository:nealbridges/VulnHunter
VulnHunter is an open-source, agentic AI security scanner that inspects source code from an attacker's perspective, confirms exploitability, and generates fixes. This maintained fork of Capital One's original project decouples the tool from Claude Code to support arbitrary agent harnesses and open-weight models.
223starsPython
Article:Jonas Reeve
Z.ai has launched GLM-5.3, keeping the GLM-5.2 base model while generating performance gains entirely through scaled-up post-training in realistic work environments. Alongside notable coding advancements, the model demonstrated an unanticipated emergent ability to chain multi-stage cyber exploits.
X post:Tibo
OpenAI has open-sourced Codex Security, a tool designed to discover, validate, and remediate vulnerabilities in codebases. It provides a CLI, a TypeScript SDK, and support for multiple model providers.
5.6kbookmarks736.6kviews
Link:OrcaRouter
OrcaRouter has released an abliterated, block-FP8 quantized version of the multimodal model Qwen3.8-27B on Hugging Face. By orthogonalising the internal refusal direction, standard safety guardrails were removed specifically for red-teaming, interpretability, and AI safety research.
Repository:bex-co/bex-security
Bex Security is an open-source, upstream-first fork of OpenAI Codex Security that brings vulnerability detection, validation, and remediation to multiple coding agents via the Agent Client Protocol (ACP).
51starsTypeScript
Parts of this content were created with the assistance of AI.
