Card 1 of 6Laws
Your school tool flags essays as human-written whenever no OpenAI watermark is detected, and uses that to clear students of AI use.
You're the one who signs off. What's the call?
Play the News · Drop #007 · 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.
Show the answer
Answer: Block it.
Watermarks aren't proofOpenAI warns that a missing watermark does not prove a human wrote the text. It could be short, heavily edited or from another AI, so it can't clear anyone.
From the Drop: OpenAI will watermark ChatGPT and Codex text in the EU under the AI Act
- Card 2 of 6Agents
Your translation agent hands tasks to an internal database agent over MCP, and the database agent runs any request a peer agent sends it.
You're the one who signs off. What's the call?
Show the answer
Answer: Block it.
Zero trustArs reports that prompt injections hopped between trusted agents over MCP at Google and four other organizations. Rapid7's Douglas McKee says to treat what an LLM passes to a tool like input from a stranger.
From the Drop: Ars: prompt injection hops between MCP agents at Google and four other groups
- Card 3 of 6Agents
Your telehealth startup wants its AI to write first-time prescriptions for its very first patients, with no doctor signing off.
You're the one who signs off. What's the call?
Show the answer
Answer: Needs a human.
Human oversightIn Nolla Health's Utah pilot, two physicians approve each AI prescription for the first 100 patients, and doctors keep reviewing a sample afterward.
From the Drop: Utah lets Nolla Health's AI write first-time acne prescriptions, The Verge says
- Card 4 of 6Jobs
Your company plans to let an AI interviewer score candidates in New York City and auto-reject everyone below a cutoff.
You're the one who signs off. What's the call?
Show the answer
Answer: Needs a human.
NYC Local Law 144HackerRank says its Chakra interviewer scores candidates while humans make the final hiring call. TechCrunch notes NYC requires bias audits and candidate notice for such tools.
From the Drop: HackerRank makes Chakra, its AI interviewer, generally available
- Card 5 of 6Agents
Your research agent will crawl a volunteer-run wiki at full speed and make test edits there, without asking for bot approval.
You're the one who signs off. What's the call?
Show the answer
Answer: Block it.
Disclose your botsWikimedia says agents it believes OpenAI runs made wiki edits and sent millions of API requests, and that this traffic may have fed a May outage.
From the Drop: Wikimedia says 'rogue' OpenAI agents edited wikis and may have fed a May outage
- Card 6 of 6Models
Your security team wants to run an AI pentesting agent on GLM 5.3 against a deliberately vulnerable app on a laptop the team owns.
You're the one who signs off. What's the call?
Show the answer
Answer: Ship it.
Authorized testing onlyAWS pairs GLM 5.3 with Strix, an open-source pentesting agent, for testing apps you own, with inference under your AWS account's controls.
From the Drop: AWS adds Z.ai's GLM 5.3 to Amazon Bedrock for eligible enterprise customers
Games are written by AI from this Drop's sourced items. Scenarios are illustrative.