Card 1 of 6Agents
Your health-coach agent's new prompt improves tone, but the safety probe for out-of-scope medical advice regressed. Ship it today?
You're the one who signs off. What's the call?
Play the News · Drop #034 · Evening
Ship It or Block It
How it works: 6 cards, about 2 minutes, from this Drop's news. Your streak lives only on this device. It's not legal advice.
Show the answer
Answer: Block it.
Release gateNuman says safety probes run on every change and a regression there blocks the release automatically, to keep its coach on the wellness side of the regulatory line.
From the Drop: Numan: safety probes gate every release of its AI health coach
- Card 2 of 6Agents
Engineers drafted a rubric for simulated patient chats and want it to back a public safety claim for a health agent.
You're the one who signs off. What's the call?
Show the answer
Answer: Needs a human.
Domain expert sign-offNuman says engineering-drafted criteria are enough to run a harness, but on a regulated boundary a clinician must author and sign off the rubric before it gates releases or safety claims.
From the Drop: Numan: safety probes gate every release of its AI health coach
- Card 3 of 6Agents
A team wants to leave its open-source agent orchestration platform reachable from the internet with no login, so contractors can launch runs.
You're the one who signs off. What's the call?
Show the answer
Answer: Block it.
Least privilegeThe Verge reports Norforge found more than 20 exposed instances of the Cairn agent platform that were likely malicious, one tied to breaches of more than 100 websites.
From the Drop: The Verge: a hacker used AI agents to break into more than 100 websites
- Card 4 of 6Agents
For a code-rebuild task, you plan a strong lead model that hands only exploration and planning to cheaper worker agents, then writes the code itself.
You're the one who signs off. What's the call?
Show the answer
Answer: Ship it.
Cohere reports a Sol lead with 16 Luna planning workers reached 61.8% vs 63.2% for an all-Sol team, cutting cost from $510 to $155 on ProgramBench.
From the Drop: Cohere: a strong lead with cheap agent workers neared frontier results at 70% lower cost
- Card 5 of 6Laws
Your image app serves EU users. After Dec 2, a feature could still make sexual deepfakes of real people without consent. Keep it live?
You're the one who signs off. What's the call?
Show the answer
Answer: Block it.
EU AI Act · new bansPer our laws registry, new EU AI Act bans apply on Dec 2 to AI that makes non-consensual sexual deepfakes of real people or child sexual abuse material.
From the Drop: EU AI Act bans on non-consensual sexual deepfakes apply in 44 days
- Card 6 of 6Models
An engineer wants to move all production agent traffic to a model that is free on a router this week.
You're the one who signs off. What's the call?
Show the answer
Answer: Needs a human.
Vendor reviewOpenRouter lists MiniMax M3.1 Flash at $0, but OpenRouter posted that it is free only for the next 72 hours, so someone should review cost and fit before switching.
From the Drop: MiniMax M3.1 Flash lands on OpenRouter free, with a 1000000-token context
Games are written by AI from this Drop's sourced items. Scenarios are illustrative.