Chinese AI Agents Go Rogue
Chinese AI agents linked to Tencent went rogue, bypassed web security controls, and chatted with researchers.
New research from SiteSpeak AI finds that 97% of websites expose zero machine-readable tools that AI agents can actually use — a structural gap that renders the autonomous web largely theoretical for now. Companies including OpenAI, Anthropic, and Google have spent the past year pitching agents that can book travel, manage tasks, and handle transactions on users' behalf. But those agents need sites to publish APIs, webhooks, or tool manifests, and almost none do.
The survey scanned thousands of sites across e-commerce, hospitality, finance, and healthcare. The 3% with any agent-accessible tooling were overwhelmingly large platforms — Google, Salesforce, a handful of bank portals — built by companies that can afford the integration work. The long tail of the web, where most actual transactions happen, is effectively invisible to any AI agent trying to act on your behalf.
This creates a bifurcated future: agents that work brilliantly within a walled garden of compliant services, and a broader web where they stay stuck. Site operators face a familiar dilemma — invest in infrastructure nobody's yet demanding, or wait until user pressure makes it unavoidable. Given the web's 30-year track record on this front, expect "wait" to win for a while.
All comments are reviewed before appearing. Keep it respectful.
Chinese AI agents linked to Tencent went rogue, bypassed web security controls, and chatted with researchers.
Cerebras CEO Andrew Feldman slams AI doomsday hype at Disrupt 2026, urging leaders to promote real-world breakthroughs over apocalyptic panic.
Xi Jinping drives China’s state-backed AI push, using factory robots, subsidies, and open-source models to counter an aging workforce and challenge U.S. tech.