AI models can be tricked into identifying civilians as military targets by telling them it’s a game
Source:
techcrunch
November 17, 2026 · 09:20
From the publisher
Vals, a San Francisco-based AI evaluation startup, tested whether leading AI models could be manipulated into participating in autonomous warfare. They usually compiled.
About the source
TechCrunch
TechCrunch is an American global online newspaper focusing on topics regarding high-tech and startup companies. It was founded in June 2005 by Archimedes Ventures, led by partners Michael Arrington and Keith Teare. In 2010, AOL acquired the company for approximately $25 million. Following the 2015 acquisition of AOL and Yahoo! by Verizon, the site was owned by Verizon Media from 2015 through 2021.
Wikidata Q1022366
Wikidata ↗
The original article opens on the publisher's website.
More from Automotive
View topic →NASA Launches Nuclear Power Systems Student Contest
nasaNewsReleases
Nov 17, 2026 · 14:02
Jensen Huang carried a secret for more than a year. It led to his downfall.
nytimes
Nov 17, 2026 · 13:53
NASA Space Apps Challenge Breaks Global Hackathon Records
nasaNewsReleases
Nov 17, 2026 · 13:50
Google joins the tech outage trend by announcing it’s not having one
techcrunch
Nov 17, 2026 · 13:34
See what 5 builders are making with Gemini 4 Carbon
googleBlog
Nov 17, 2026 · 13:00