Reflection AI lets its Beam agent run a real vending machine to probe autonomy risk
alexpolozov · x · 2026-11-22
Reflection AI partnered with eval lab Andon Labs to install independently operated vending machines ("Mr. Beam") in its NYC and SF offices, testing the Beam model's autonomy and loss-of-control risk in a live deployment with no human intervention.
Early findings:
- Strong judgment and resistance to direct jailbreaks
- Struggled to consistently execute multi-step workflows
- Remained susceptible to fabricated claims that shaped later actions
Reflection argues autonomous safety measurement is about mapping agent reliability boundaries, not tallying jailbreak failures, and is refining how it evaluates tool calls and memory updates as Beam expands into a more complex multi-actor internal economy.
More from Models
- Andrew Ng: Gemini Neon is the only model that doesn't talk like an insufferable AI — andrewgwils · 2026-11-22
- Bindu Reddy: Neon high matches Opus high quality but runs much slower — bindureddy · 2026-11-22
- Claude Max/Ultra users get $100–$200/mo in API credits most never claim — here's how to actually use them — jekapats · 2026-11-22
- TensorFold 1.3.3 ships 124 PRs in 24h, boosts GLM decode to 172 tok/s on GB10s — AIFlow_ML · 2026-11-22
- SO-Arena: NUS benchmark separates capability and decision reliability for single-pass models — 新智元 · 2026-11-22
- OpenAI Sweeps Top 7 Spots on Updated Arena Alignment Index; Kimi K3.1 Pro Best Open Model — gdb · 2026-11-22