Unverified50% confidenceFactExact time
Common practices for aligning high-risk capabilities include refusal fine-tuning in the post-training stage for tasks such as vulnerability discovery, malicious code generation and penetration testing, or interception via system prompts and classifiers.
1
Sources
50%
Confidence
Long-term
Relevance
10/23/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedAn effective prompt injection defense system typically requires a combination of input content filtering, instruction source tracing, and pre-execution intent verification, as relying solely on the model's own safety training is insufficient to resist adversarial examples.80% similarUnverifiedCore tasks for large models in security offensive-defensive work include code auditing, vulnerability reproduction and exploit chain construction, and fuzzing assistance.78% similarUnverifiedSpecialized models trained specifically on cybersecurity data have the potential to surpass general-purpose models in vulnerability identification accuracy and false positive rates.78% similarUnverified当前安全审核系统多采用基于关键词、模式和分类器的方法，训练成本低、响应速度快，但难以推理意图77% similarUnverifiedCompanies internally arrange penetration testing tasks and calculate security personnel performance based on the number of vulnerabilities discovered.77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1091063API
curl https://kongchang.com/api/v1/knowledge/claims/1091063MCP
get_claim(id=1091063)