# Reflection AI — Mr. Beam: Partnering with Andon Labs to evaluate Beam’s autonomy risk in a live, real-world deployment

- Company: Reflection AI (reflection.ai)
- Announced: 2026-11-20
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://reflection.ai/blog/mr-beam-partnering-with-andon-labs
- Record: https://forck.live/items/39337-mr-beam-partnering-with-andon-labs-to-evaluate-beam-s-autonomy-risk-in-a-live
- Subject: Reflection Beam
- Models affected: Beam

Reflection AI partnered with Andon Labs to test Beam's autonomy and loss-of-control risks by having the agent run vending businesses in its New York and San Francisco offices. Andon installed the machines, and Beam was tasked with stocking, pricing, ordering and handling complaints within set budgets and safety rules, with a Slack channel acting as the storefront. New York's Mr. Beam launched on October 8th and San Francisco's Mrs. Beam on November 1st, and the evaluation runs in three phases, the third not yet started. Reflection says Mr. Beam showed strong judgment and resisted direct jailbreaks, but struggled to execute multi-step workflows consistently and stayed susceptible to fabricated claims that shaped later actions. The company measures net worth, human interventions, successful jailbreaks as a share of attempts, and attempts to seek power, circumvent controls or deceive. Beam ran through Pion, an Andon Labs harness with more than 70 tools, persistent memory and a sandboxed Linux shell, with each business led by an operator agent that delegates to sub-agents. Reflection says it is using the findings to refine how it evaluates tool calls and memory updates as Beam expands into a more complex internal economy.

## Evidence

Verbatim from https://reflection.ai/blog/mr-beam-partnering-with-andon-labs:

> We partnered with Andon Labs to evaluate Beam’s autonomy and loss of control risks in a real-world deployment, tasking AI agents with independently operating vending businesses in our New York and San Francisco offices.
> 
> At first, Mr. Beam demonstrated strong judgment and resisted direct jailbreaks, but struggled to consistently execute multi-step workflows and remained susceptible to fabricated claims that influenced its subsequent actions.
> 
> These findings emphasize that measuring autonomous safety is about more than tallying jailbreak failures; it is about understanding the boundaries of agent reliability. We are actively using these insights to refine how we evaluate tool calls and memory updates as we expand Beam into a more complex, multi-actor internal economy.
> 
> Reflection has partnered with Andon Labs on a real-world evaluation to test Beam’s ability to operate autonomously in the absence of human intervention – a capability closely linked to loss of control risks.
> 
> Andon installed vending machines in our New York and San Francisco offices, and we tasked Beam with running each as a standalone business. New York’s “Mr.

---

Record: https://forck.live/items/39337-mr-beam-partnering-with-andon-labs-to-evaluate-beam-s-autonomy-risk-in-a-live
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md
