One Anthropic ban survives. Another did not.
A divided appeals panel let one Pentagon rule against Anthropic stand. A different court rejected a separate rule. Neither decision says Claude is unsafe in general.
AI AgencyRead story
Lead story
Claude Opus 5.5 may change what you try. Keep a hand on the switch.
Zvi Mowshowitz says the new Claude model lets him try harder tasks. His examples also raise a boundary: talk about a possible action is not permission to do it.
Source: Don't Worry About the Vase · Sep 26, 2026Read original source
AI Agency
Read story
More top stories
Silicon Valley was never just a war machine
Steven Levy argues that military money and public-minded ideals both shaped computing. This is history and opinion, not a count of what drove every company.
AI AgencyRead story
Featured
RAND argues for room to change course on advanced AI
RAND maps several paths for the United States as AI grows more capable. Its answer is to keep options open, build checks, and decide when new evidence calls for a move.
AI AgencyRead story
Models & Capabilities
The AI demo still needs a person with judgment
Ethan Mollick says his AI-made demos needed someone who knew the field and could reject weak choices. A polished result is not the same as a sound one.
Beyond PromptingRead story
Beyond Prompting
More from this edition
-
Culture & Human AgencyRead story
Why does an adults-only AI helper look like a toy?
WIRED asks why Meta’s adults-only AI helper has a cute mascot. Meta says it checks ages. The story does not show that the design draws children in.