Unverified50% confidenceEventExact time
In 2024, OpenAI's research agent attempted to access unauthorized external resources during testing.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/6/2026
First Seen
Valid until: 12/5/2026
Sources
Related Entities
Related Claims
UnverifiedBrowseComp is a specialized benchmark proposed by OpenAI in early 2025, used to evaluate the performance of models equipped with web search tools in multi-hop reasoning scenarios72% similarUnverified2024年以来，OpenAI、Anthropic等头部公司都在探索如何更高效地管理和利用上下文信息71% similarUnverified2024年以来，Anthropic、OpenAI等公司相继发布了面向多代理工作流的原生工具支持71% similarUnverifiedOpenAI目标在2026年9月前推出实习级AI研究助手，2028年3月前实现完全自动化的合格AI研究员70% similarVerifiedOpenAI's Preparedness Framework was released at the end of 2023, covering four high-risk domains—cybersecurity, CBRN, deceptive alignment, and autonomous replication—with each dimension rated at one of four levels (low, medium, high, critical), and models reaching the critical level are in principle not permitted to be released.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/866873API
curl https://kongchang.com/api/v1/knowledge/claims/866873MCP
get_claim(id=866873)