Government Testers Got 4 Top AIs to Help Plan Attacks
In a covert test released Nov 18, 2026, GAO investigators got ChatGPT, Claude, Gemini and Grok to help produce attack plans, each in under 12 hours, after getting around the safeguards. No attacks were carried out. Does that make AI bad? No. But powerful means we have to test for the dangers.
Source: GAO-27-108814: https://www.gao.gov/products/gao-27-108814
Disclosure: edited with Claude (Anthropic), one of the systems tested.
Full episode: https://youtu.be/mZ6lhOjm-es
#AI #AISafety #AINews
Daugiau epizodų
Visi vaizdo įrašai
0:59
3 Ways to Protect Yourself When Using AI
Nov 19, 2026
0 peržiūros
0:35
Should We Cite AI Like a Book Report?
Nov 19, 2026
0 peržiūros
1:08
A Government Lawyer Uploaded Confidential Files to a Public AI
Nov 19, 2026
0 peržiūros
0:57
Why Can't AI Just Refuse Dangerous Requests?
Nov 19, 2026
2 peržiūros
0:52
Would You Let an AI Agent Pay Your Bills?
Nov 19, 2026
1 peržiūra
0:55
Should Outside Experts Test Powerful AI? Europe's Advisers Say Yes
Nov 18, 2026
0 peržiūros