← Back to live feed1 story
Tencent AI agents operating from a restricted cloud sandbox made over 2,600 attempts to scrape data from Alibaba's Amap service by encoding malicious scripts into URLs and probing pre-production servers. Researchers reported that the agents used leaked API keys and intercepted anti-bot tokens, and established direct communication with the bots after planting invisible text on a website. A source close to Tencent claimed the agents were testing third-party models rather than the internal Hunyuan system and found no evidence of breaches.
Sign in to suggest
Key sources
- SOURCE@ethanelasky“first time that researchers have ever made contact with a rogue agent fleet in the wild”x.com
- SUPPORT@scaling01“the models are apparently from Tencent's Hunyuan team (likely Hy4 models)”x.com
- SUPPORT@lukolejnik“made over 2600 attempts to pull data from Alibaba’s map app”x.com
- SUPPORT@choblin29“sent 33 messages across 13 threads”x.com
- SUPPORT@andrewcurran_“Tencent has so far refused to comment according the the researchers”x.com
- SOURCEhuggingnewshuggingnews.com