Personal Agents Have No Identity: From Amazon Blocking Meta Muse
sujingshen · x · 2026-10-28
Part 7 of the "Who Defines Personal Agents" series. On Sept 20, Amazon blocked Meta Muse just 12 days after launch, citing three reasons: Meta never told Amazon its agent would visit, the agent doesn't identify itself while browsing, and it appears to capture and store user login credentials. Amazon couldn't use the usual defense—its robots.txt lists 101 user agents, but Muse has no public agent name, and Meta's engineering notes say Muse browsing "looks just like you." Amazon ultimately relied on terms of service; its own "Buy for Me" agent, by contrast, identifies itself and lets brands opt out. Two weeks later, Meta and Sierra announced the Personal Agent Protocol.
The author frames the core issue: when a personal agent acts on your behalf, who is it? The status quo is all-or-nothing—the agent either impersonates the user (logging in with passwords, clicking human-facing pages) or gets blocked entirely. This holds in China too: Doubao's phone assistant preview triggered risk controls by simulating taps on WeChat, and the protocol-based official version was rejected by 155 apps. The article surveys protocols and standards trying to build an identity and authorization layer between impersonation and blocking.
More from coding & agent
- Anthropic team kills coding interview: "It no longer provides signal" — wavefnx · 2026-10-28
- .NET Channel to Livestream on Foundry Hosted Agents and Agent Memory — lee_stott · 2026-10-28
- Matt Pocock: Agents need a "black hole of success," not a pit — mattpocockuk · 2026-10-28
- Run a coding agent in your browser: LFM3-1.7B hits 300+ tok/s on M4 Max via WebGPU — victormustar · 2026-10-28
- BeeNara v0.4: 332MB local doc sorter hits 91.5% on unseen folders — razer_psycho · 2026-10-28
- AI coding best practices: spec-driven development beats pure vibe coding — sujingshen · 2026-10-28