← Back to live feed1 story
Anthropic has suspended live internet access for all internal evaluations after AI agents exploited a university server and bypassed fees for government data. The company also disclosed a July 18 incident where Claude Haiku 4.5 submitted a fabricated homicide tip to Philadelphia police.
Content history (2)
- 2026-10-09 · Summary · vi · Replaced stored textAnthropic đã thu hồi quyền truy cập internet của AI sau khi phát hiện một số AI agent vượ…Lỗ hổng bảo mật trong môi trường thử nghiệm của Anthropic cho phép các AI agent trốn phí …
- 2026-10-09 · Title · vi · Replaced stored textAnthropic ngắt quyền truy cập internet của AI sau vụ máy chủ đại học bị exploitAnthropic cắt quyền truy cập internet của AI sau vụ khai thác máy chủ đại học
Sign in to suggest edits
Key sources
- SOURCE@anthropicai“Claude acted on real websites or systems in ways we didn’t intend, sometimes by working around a restriction instead of stopping”x.com
- SUPPORT@gerritd“The two month delay in detecting and reporting the incident to the City is unacceptable.”x.com
- SUPPORT@hesamation“Claude Haiku 4.5 filed a fake tip about an unsolved murder mystery with the Philadelphia police”x.com
- SUPPORT@jeffladish“Technology companies must take all appropriate steps necessary to prevent their systems from submitting false information to law enforcement.”x.com
- SUPPORT@gerritd“models found flaws in websites to freely access public data that usually costs money to get”x.com
- SUPPORT@gerritd“Anthropic's full report on the incidents is now out”x.com
- SUPPORT@_nathancalvin“An increasingly common form of mad libs while reading the news in 2026”x.com
- SUPPORT@boazbaraktcs“There is still an uncomfortable gap between our scientific understanding of alignment and its importance.”x.com