Play the News · Drop #013 · Morning

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.

Card 1 of 6Agents
Your lending app wants Claude to approve or decline small loans automatically, with no staff member reviewing its recommendations. Ship it?
You're the one who signs off. What's the call?
Show the answer
Answer: Needs a human.
Anthropic Usage Policy · high-risk
Anthropic's Usage Policy requires a qualified human in the loop, someone who can review and change Claude's recommendations, for high-risk uses like finance.
From the Drop: Anthropic updates its Usage Policy and bans sustained abuse of Claude
Anthropic · Oct 8 The Verge · Oct 8
Card 2 of 6Agents
A warehouse team connects Claude to a robotic arm that moves heavy pallets. Nobody on site can observe or stop the arm while it runs.
You're the one who signs off. What's the call?
Show the answer
Answer: Block it.
Anthropic Usage Policy · hardware
Anthropic's updated policy requires that Claude-driven hardware can be stopped. With no operator able to halt the equipment, this setup shouldn't go live.
From the Drop: Anthropic updates its Usage Policy and bans sustained abuse of Claude
Anthropic · Oct 8 The Verge · Oct 8
Card 3 of 6Agents
Your open-source project enrolls in OSS Scanner and plans a bot that posts every incoming report straight to the public issue tracker.
You're the one who signs off. What's the call?
Show the answer
Answer: Needs a human.
Triage before disclosure
Anthropic says OSS Scanner reports are sent without human review and expects a true-positive rate above 90%, so some will be wrong. A maintainer should triage first.
From the Drop: Anthropic launches a Cyber Mission with free OSS Scanner security scans
Anthropic · Oct 8 The Verge · Oct 8
Card 4 of 6Agents
An agent pays a few cents per model call from a customer-funded wallet, and the spending cap is enforced by the payment infrastructure, not the prompt.
You're the one who signs off. What's the call?
Show the answer
Answer: Ship it.
Least privilege
AWS says AgentCore payments enforces session limits outside the model, so a manipulated prompt can't push the agent past the customer's cap.
From the Drop: AWS shows agents paying per inference under spending caps set outside the model
AWS Machine Learning Blog · Oct 8
Card 5 of 6Agents
A monitoring probe flags one of your hosted agent's sessions as possible malicious hacking. Your config can log it, refuse, or route it to a person.
You're the one who signs off. What's the call?
Show the answer
Answer: Needs a human.
Escalate flagged sessions
TechCrunch says Goodfire's probes caught 93% of malicious hacking sessions in tests, and Baseten customers can send flagged events for human review.
From the Drop: Goodfire launches probes that watch agents from inside the model, on Baseten
TechCrunch · Oct 8
Card 6 of 6Agents
Your company gives an enterprise agent its own Workspace account to schedule team meetings, with every action logged under the agent's name.
You're the one who signs off. What's the call?
Show the answer
Answer: Ship it.
Auditability
Google says its Gemini agent gets its own account and writes an audit trail attributed to the agent rather than a person, so its actions stay traceable.
From the Drop: Google's Gemini agent gets its own Workspace account and audit trail, TechCrunch says
TechCrunch · Oct 8 The Verge · Oct 8

Games are written by AI from this Drop's sourced items. Scenarios are illustrative.