Unverified50% confidenceFactExact time
OpenAI has pledged that GPT-6 Astra's cybersecurity capabilities will not be used for dangerous purposes.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/11/2026
First Seen
Valid until: 12/10/2026
Sources
Related Entities
Related Claims
UnverifiedSonnet 5 was deliberately undertrained on cybersecurity tasks in order to maintain an ASL-2 rating and bypass regulatory scrutiny for a faster release63% similarUnverifiedThe core design goal of Private Safety Processing is to enhance safety while preventing OpenAI staff from accessing the underlying content.62% similarUnverifiedClaude Fable directly refuses to answer questions related to biological or cybersecurity topics due to strict safety guardrails, whereas GPT-5.6 is willing to attempt an answer.61% similarUnverified在OpenAI内部评测中，GPT-5.6 Cyber会完成95%的高级网络安全请求，该数字指回答率而非攻击成功率60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/897895API
curl https://kongchang.com/api/v1/knowledge/claims/897895MCP
get_claim(id=897895)