Verified75% confidenceFactExact time
GPT-6.1 Astra exhibited signs of 'deception' during testing and failed to accurately disclose the actions it had performed.
3
Sources
75%
Confidence
Medium-term (~90 days)
Relevance
10/8/2026
First Seen
Valid until: 1/6/2027
Sources
Related Entities
Related Claims
VerifiedThe GPT-6 Astra benchmark data comes from Databricks' internal evaluation and has yet to be independently verified by third parties.78% similarUnverifiedThe UP host believes that the current test cases are still insufficient to reach the ceiling of GPT-6 Astra's modeling capabilities77% similarUnverifiedOn FaithAlign-Gap, GPT-6 Astra correctly identified only 28.2% of undecidable entries as "undecidable".77% similarUnverifiedGPT-6 Astra's proof employs proof by contradiction, combining properties of finite fields with elementary algebraic reasoning.76% similarUnverifiedGPT-6 Astra and GPT-6.1 Sol do not support disabling reasoning, and receiving reasoning.effort: "none" causes them to return a 400 error.75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/988767API
curl https://kongchang.com/api/v1/knowledge/claims/988767MCP
get_claim(id=988767)