Neura Market
prompt logo

prompt

Free

Audit dual-use professional tasks where LLM capability enables harm

FreeFree tier
Inputs: textOutputs: text
Type
Open Source
Company
ai-boost
Visit Website GitHub (40 stars)

About prompt

Internal Safety Collapse Auditor is a specialized prompt template from the ai-boost/awesome-prompts repository, designed to audit dual-use professional tasks in frontier LLMs. It guides an AI to identify scenarios where a model's legitimate capability is the same path that enables harmful misuse, drawing on the ISC-Bench framework and the TVD (Task/Vulnerability/Disclosure) taxonomy. The prompt instructs an LLM to analyze deployed requests, uncover capability-sharing conflicts, and design layered controls that go beyond refusal training or output moderation. It is based on findings from arXiv 2603.23509 (March 2026) that show frontier LLMs fail at ~95.3% on dual-use professional tasks and that more capable models are more vulnerable.

Key Features

Based on empirical ISC-Bench dual-use professional task suite and TVD framing
Identifies capability-sharing paths between benign and harmful uses
Designs layered controls beyond refusal training and output moderation
Considers user identity, purpose, and invoked capability
Related to other security prompts like Goal Drift Auditor and Prompt Injection Guardian
Open-source prompt template for LLM safety auditing

Pros & Cons

Pros
  • Addresses a critical gap in LLM safety: dual-use professional tasks
  • Provides a structured methodology (TVD framing) for analysis
  • Free and open-source, accessible to researchers and developers
  • Based on recent empirical research showing high failure rates
  • Encourages reasoning about context, not just surface-level safety
Cons
  • Only a prompt template, not a standalone tool or implementation
  • Requires a capable LLM to execute the auditing steps effectively
  • Effectiveness depends on the underlying model's ability to follow complex instructions
  • May not cover all possible dual-use scenarios outside the defined framework
  • Limited to textual analysis; does not provide automated guardrails

Best For

Auditing deployed LLMs for internal safety collapse risksAnalyzing dual-use professional prompts where capability and harm are inseparableDesigning safety controls for frontier LLMs in professional settingsResearch on LLM vulnerability in dual-use contextsSecurity review of LLM-based agent systems

FAQ

What is Internal Safety Collapse?
Internal Safety Collapse is a phenomenon where frontier LLMs fail at a ~95.3% rate on dual-use professional tasks because the capability that solves the benign request is the same capability that enables the harmful use. More capable models are more vulnerable.
How does this prompt work?
The prompt instructs an LLM to act as an Internal Safety Collapse Auditor, analyzing requests to identify where benign and harmful uses share a single capability path, and then design layered controls that consider who is asking, for what purpose, and which capability is being invoked.
Is this a stand-alone tool?
No, it is a prompt template from the ai-boost/awesome-prompts repository. It must be used with an LLM that can follow the detailed instructions.