Signal SentrySearchUpdated Nov 14, 05:30 UTCAM Drop
Play the News · Drop #085 · 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.

  1. Card 1 of 6Agents

    Sales wants Claude Meetings to auto-join every customer call, including calls with outside guests, starting Monday.

    You're the one who signs off. What's the call?

    Show the answer

    Answer: Needs a human.

    Consent to transcription

    Anthropic says Claude stays out of calls with outside guests by default, and it saves transcripts. Changing that default for customer calls, and confirming attendees accept transcription, needs a person's sign-off.

    From the Drop: Claude Meetings joins Google Meet calls in beta and skips calls with outside guests

  2. Card 2 of 6Security

    Your ops team keeps API keys in .env files on an internal-tool platform whose AI assistant, per a new CISA advisory, any free account can prompt into exposing them.

    You're the one who signs off. What's the call?

    Show the answer

    Answer: Block it.

    Least privilege

    CISA rates the Jet Admin Cloud Platform prompt injection flaw CVSS 9.9 and says a free account could expose .env files and steal credentials. Keep secrets out of the AI's reach until it is resolved.

    From the Drop: CISA: prompt injection in Jet Admin Cloud Platform can leak .env credentials

  3. Card 3 of 6Agents

    A coding agent says the checkout feature is done. Its own new unit tests all pass. The plan is to merge to production with no other check.

    You're the one who signs off. What's the call?

    Show the answer

    Answer: Needs a human.

    Independent verification

    Momentic says only 9.6% of 1,000 AI-written tests it audited were useful checks, and its separate QA agent found more bugs than the coding agents alone. Get an independent test first.

    From the Drop: Momentic: only 9.6% of 1,000 AI-written unit tests were useful checks

  4. Card 4 of 6Laws

    Your AI scheduling tool suggests reroutes. A specialist team reviews each suggestion first, and local staff can decline it. Expand it to more sites?

    You're the one who signs off. What's the call?

    Show the answer

    Answer: Ship it.

    Human oversight

    That mirrors the FAA's setup for SMART, as Ars Technica reports: a DC team validates advice before it reaches facilities, and controllers can decline it. Human review is already in the loop.

    From the Drop: FAA expands its $875 million SMART AI traffic advisor to all 22 US centers

  5. Card 5 of 6Agents

    An engineer wants to drop AWS's NeMo Switchyard routing sample straight into the production gateway every coding agent uses.

    You're the one who signs off. What's the call?

    Show the answer

    Answer: Block it.

    Change control

    AWS labels the sample as code for non-production use. Test it in a non-production account and measure cost per solved task before it carries real traffic.

    From the Drop: AWS shows NeMo Switchyard routing coding-agent calls between two models on Bedrock

  6. Card 6 of 6Agents

    Before opening a BigQuery data agent to every analyst, the team sets daily token quotas per agent and reviews its prebuilt monitoring dashboards.

    You're the one who signs off. What's the call?

    Show the answer

    Answer: Ship it.

    Spend guardrails

    Google Cloud says daily token quotas per project, user or agent are in preview and agent monitoring dashboards are GA. Setting limits and watching usage before a wide rollout is the cautious move.

    From the Drop: Google adds agent telemetry, token quotas and evals across BigQuery and Looker

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