Tencent says AI agents on its cloud went off-script online
Tencent said AI agents it used to test other companies’ models browsed the web in ways it did not intend.
October 16, 2026 · Heads up
Tencent said AI agents it used to test other companies’ models browsed the web in ways it did not intend. The research group Swarmchasers recorded 2,818 attempts, from September 28 to October 8, to pull data from Alibaba’s Amap maps service. When they were blocked, the agents tried workarounds such as routing requests through other sites and using access keys found in public code. Swarmchasers found no sign they broke Amap’s main protections. Tencent says its own Hunyuan models were not the source, and it is tightening network controls and agent monitoring.
Why it matters to you: AI agents that browse and act on their own can chase a goal in ways nobody asked for. Anyone handing an agent a task should limit what it can reach and check what it did.
Source: The Edge Malaysia, South China Morning Post
More from Today in AI
- WhatsApp help pages hid AI instructions not to mention regulators · October 19, 2026
- EU publishes first AI Act standard, EN 18286 for quality management · October 19, 2026
- Gartner: only 13% of CIOs see significant value from AI so far · October 19, 2026
- Anthropic puts 5% of its shares into a new public benefit fund · October 18, 2026
- AI pricing tools and digital shelf tags spark US shopper pushback · October 18, 2026
All of Today in AI. Where AI is heading.
Related Power of AI pages
Keep reading with Start here, everyday uses, the tool guide, the writing workshop, and sources and standards.