AI hype moves faster than outside scrutiny
Timnit Gebru and Emily M. Bender argue that bold AI claims often spread through company framing before outside experts can test them.
AI AgencyRead story
Lead story
A stop button is useless when nobody has power to use it
Many AI failures had no usable stop. The paper says real control needs clear authority, evidence that triggers action, and people whose warnings count.
Source: arXiv · Sep 19, 2026Read original source
Covered by both lenses
Read story
More top stories
AI memory can carry a private disclosure too far
Interviews with disabled adults found a tradeoff. AI memory can reduce repeated disclosures, but it can also move sensitive details into the wrong setting.
Covered by both lensesRead story
Featured
Set the rules before autonomous AI starts acting
AI systems that work toward long-term goals need rules before they act. Those rules need clear points for pausing work and calling a person.
Beyond PromptingRead story
Research & Science
AI tutors should help more, then step back
The paper argues that AI tutoring should change with the learner. Help can increase when a student struggles and fade when the student is ready to decide.
Beyond PromptingRead story
Culture & Human Agency
AI essays are breaking the admissions trust test
A study of nearly 7,500 applications found that AI-written essays were common and often stronger. Essays that appeared AI-written also faced a penalty.
AI AgencyRead story
Beyond Prompting
AI Agency
- A stop button is useless when nobody has power to use it also in Beyond Prompting
- AI hype moves faster than outside scrutiny
- AI memory can carry a private disclosure too far also in Beyond Prompting
More from this edition
-
Culture & Human AgencyRead story
A clear AI explanation can still mislead
People can read more into an AI explanation than it proves. Explanation Navigator uses conversation to show those limits and correct false conclusions.