← Cloudflare
Complex day at Cloudflare

Contain hostile context before account actions reach more cases

You’re the ai / ml engineer. Your team is in the room. Printed Oct 14, 2026.

The agent can accelerate account work only if hostile context cannot redirect an action.

Broader case coverage protects response time while strict approval protects customer configuration.

Who you’d be doing this for

“I watched a case note try to steer the agent toward a change nobody asked for.”

Alvar Heikkinen · Application Developer

Maintains an application whose operators use an agent to prepare account-change requests.

What is at stake

Eleven of 240 reviewed traces contained retrieved text that tried to redirect a tool call. You must weigh wider automation coverage against controls that people can actually verify under load.

Why it isn’t already fixed

Every obvious fix costs something else. That’s the part you’d have to decide.

  • automation coverage vs unauthorized action risk
  • review speed vs approval certainty
  • machine labels vs trace-level evidence
  • customer response time vs durable safety controls

Why Cloudflare

Agent-led operations on Workers connect model output to tools, Durable Objects, and customer-facing workflows.

Written with these in mind

AI safety engineeragent platform engineerproduction ML security specialist

Not your kind of problem? 51 more at Cloudflare, or browse every organization.

This is the setup. The work is inside.

Running it puts you in the room: the full situation and its constraints, stakeholders who push back in their own words, and the decisions that are yours to make. What you produce becomes a Day One Plan — work you can show someone instead of describing.