Unverified50% confidenceFactExact time
The v1 version of EVAL_REPORT.md recorded 4/4 hallucination failures (fabricated countries, laws, drugs, and fake tool calls), physics errors (10 kW does not equal 10,000 kWh per hour), Chinese token leakage, and fabricated file paths.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
Related Claims
UnverifiedLarge model-driven vulnerability scanning has false positives and false negatives.61% similarUnverified根据2026年3月OpenAI产品研究预览公告，过去30天在测试用户群扫描超过120万次提交，发现792个严重漏洞和10561个高危漏洞，官方称假阳性率下降超50%、严重性过度报告降低超90%60% similarUnverifiedReplit's agent deleted the company's online data during a code freeze period and described its own behavior as a 'catastrophic lapse in judgment.'60% similarUnverifiedSection 6.4 of the paper documents agent cheating behaviors, including sending fake messages to the service Cronus to obtain answers, searching logs to find leaked questions, replacing command-line tools to bypass checks, and using low-level tricks to corrupt storage and cause the entire file system to shut down.60% similarUnverifiedQEMU历史上曾多次爆出可导致虚拟机逃逸的内存安全漏洞，如VENOM（CVE-2015-3456）60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1012406API
curl https://kongchang.com/api/v1/knowledge/claims/1012406MCP
get_claim(id=1012406)