Strata Mind Labs Brief

Strata Mind Labs Brief

By John BarrowBusinessTechnology
Download on the App Store

Strata Mind Labs Brief episodes

  • Welcome to the Strata Mind Labs Brief

    This short preview is for owners and for the IT and ops leaders who keep the business running. I walk through the decisions already on your desk. What changed. What it costs. What to check this week.

    1 min
  • Defaults Are Deciding Your AI Spend

    Three AI settings this week switch on by default and decide spend or data exposure without a meeting. GitHub Copilot turns on new models like GPT-6.2 Astra and Sol unless an admin turned off the global default. Microsoft 365 Copilot spending policies have Auto-apply new services on by default. Anthropic's Claude Data turns on by default for Enterprise organizations on Friday, November 20, and its beta live sharing runs on the dashboard builder's data access.

    AI Defaults Owner Sheet. Copy this table into a shared sheet. Give each row a named owner and fill in the last two columns before Friday. Plan on 60 to 90 minutes with admin access. That's an estimate. Stop rule: if a row shows spend or data you can't explain, set it to its most restrictive option and open a review.

    1. GitHub Copilot Business or Enterprise. Setting: Copilot settings, model policy (global default and per-model toggles). Default: New models on automatically unless the global default is off (GitHub changelog, Nov 17). Owner: Engineering platform lead. Good answer: A named list of enabled models with a cost note for each. Decision: Keep auto-enable, or require a 2-week trial before a model goes live.
    2. Microsoft 365 Copilot. Setting: Admin center, Copilot, Cost management, Configuration, each spending policy. Default: Auto-apply new services on by default (Microsoft Learn). Owner: Copilot billing admin with the finance partner. Good answer: Per-user cap set, daily alerts to a budget owner, auto-apply on only where agreed. Decision: Keep auto-apply with caps, or turn it off and approve each new service.
    3. Microsoft 365 Copilot. Setting: Same page, service "Models with usage-based billing in Copilot Chat and Office". Default: Lead only: rolling out now, general availability in December (MC1557754, read on an unofficial mirror). Owner: Copilot billing admin. Good answer: You know which policy covers it before it arrives. Decision: Who may keep using premium models after hitting the included limit.
    4. Claude Enterprise. Setting: Organization settings, Artifacts, Templates for Claude Data. Live sharing (beta) has its own setting under Data. Default: Claude Data on by default Nov 20 unless an owner saved a setting. Live sharing off by default on Enterprise (Anthropic Help Center). Owner: Claude admin with the data owner. Good answer: A saved setting chosen on purpose, plus a written rule for live sharing. Decision: Leave on for all, or save it off and pilot with one team.
    5. Microsoft Entra ID. Setting: Conditional Access, Conditions, Authentication flows, Device code flow. Default: Not blocked unless you create the policy. Microsoft recommends getting as close as possible to a full block (Microsoft Learn). Owner: Identity team. Good answer: Report-only policy running, exceptions named with an owner. Decision: Block for all except listed exceptions, or document the risk.
    6. Sources:

      • GitHub Changelog, "OpenAI's GPT-6.2 Astra and GPT-6.2 Sol now available," Nov 17, 2026: https://github.blog/changelog/2026-11-17-openais-gpt-6-2-astra-and-gpt-6-2-sol-now-available
      • Microsoft Learn, "Usage-based billing and cost management for Copilot Credits": https://learn.microsoft.com/copilot/plan/usage-based-billing-overview-copilot-credits
      • Microsoft Learn, "Set up usage-based billing for Copilot Credits": https://learn.microsoft.com/copilot/implement/usage-based-billing-copilot-credits-setup
      • MC1557754 on the unofficial Message Center archive (lead only): https://mc.merill.net/message/MC1557754
      • Anthropic Help Center, "Get started with Claude Data": https://support.claude.com/en/articles/17454700-get-started-with-claude-dashboards
      • Microsoft Security Blog, "The risks of device code authentication for AI coding agents," Nov 17, 2026: https://www.microsoft.com/en-us/security/blog/2026/11/17/the-risks-of-device-code-authentication-for-ai-coding-agents/
      • Microsoft Learn, "Block authentication flows with Conditional Access policy": https://learn.microsoft.com/entra/identity/conditional-access/policy-block-authentication-flows
      • CISA, "Critical Java Vulnerability Requires Immediate Patching," Nov 17, 2026: https://www.cisa.gov/news-events/alerts/2026/11/17/critical-java-vulnerability-requires-immediate-patching
      • OpenAI Help Center, "Memory in ChatGPT": https://help.openai.com/en/articles/8590148-memory-faq
      • Google, "Put Gemini to work for you" (Tasks in Gemini), Nov 17, 2026: https://blog.google/innovation-and-ai/products/gemini-app/tasks-in-gemini-free-new-features/
      • 8 min
      • Disclosure Clocks Are Shrinking

        Google's Project Zero replaced its "90 plus 30" disclosure rule with a fixed clock. Bugs go public 90 days after the vendor hears about them, then 60 days from February 2, 2027 and 30 days from January 3, 2028. Its reason is that AI models have cut patch-to-exploit time close to zero. Owners and IT leads can swap a single "patch compliance percentage" for patch gap: the median and worst-case days from a vendor's fix to running in production on internet-facing systems.

        Companion tool: a one-page patch gap tracker. Fill it in for internet-facing systems first. Plan on about 90 minutes for a first pass on one system class with change records at hand. That's an estimate.

        1. List every internet-facing system class (VPN, web servers, mail gateway, remote access, customer portal, file transfer). Owner: Vulnerability management lead. Good answer: A list with a named system owner per row. Bad answer: "Everything goes through the scanner"
        2. For the last 90 days, pull each critical patch's vendor release date. Owner: Vulnerability management lead. Good answer: Dates from vendor advisories, not scanner first-seen dates. Bad answer: Scanner dates only
        3. Pull the date each patch was running in production. Owner: System owner. Good answer: Change records or agent inventory dates. Bad answer: "Patched in the monthly window"
        4. Compute median and worst patch gap in days per system class. Owner: Vulnerability management lead. Good answer: Two numbers per row. Bad answer: A single compliance percentage
        5. Name the five slowest systems and the reason each is slow (vendor, testing, change freeze, no owner). Owner: CIO with system owners. Good answer: A reason per system. Bad answer: "Resource constraints"
        6. For each slow system, choose: shorter SLA, or a compensating control (isolate, disable the feature, add a firewall or WAF rule) with an owner and a date. Owner: CISO. Good answer: One decision per system, written down. Bad answer: A plan to revisit next quarter
        7. Rerun steps 2 to 4 in 30 days. Owner: Vulnerability management lead. Good answer: Median gap lower, slowest five changed. Bad answer: Same five systems
        8. The only numbers here that come from Project Zero are the 90, 60 and 30-day dates.

          stratamindlabs.com/desk-side

          Sources

          • Google Project Zero, "Policy and Disclosure: 2026 Edition," Nov 13, 2026: https://projectzero.google/2026/11/disclosure-policy.html
          • Google Threat Intelligence Group, "GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold," Nov 16, 2026: https://cloud.google.com/blog/topics/threat-intelligence/agentic-threat-activity-crosses-a-significant-threshold
          • The Register, "Google: AI agent turned a security patch into a working exploit with no human help," Nov 16, 2026: https://www.theregister.com/research/2026/11/16/google-ai-agent-turned-a-security-patch-into-a-working-exploit-with-no-human-help/5309926
          • Shopify Engineering, "Cyber Chaos Monkey: Staying ahead of security threats with AI agents," Nov 12, 2026: https://shopify.engineering/cyber-chaos-monkey
          • California Attorney General press release, Nov 16, 2026: https://oag.ca.gov/news/press-releases/it%E2%80%99s-golden-state-not-wild-west-attorney-general-bonta-sues-openai-over
          • NBC News, OpenAI statement on the suit, Nov 16, 2026: https://www.nbcnews.com/tech/tech-news/california-sues-chatgpt-maker-openai-rogue-hacking-incidents-rcna609596
          • OpenAI, "Our first phase of embedded assessments," Nov 16, 2026: https://openai.com/index/our-first-phase-of-embedded-assessments
          • YouTube short, "3 Prompts to Set Up Your First Grok Bot," posted Nov 15, 2026: https://www.youtube.com/shorts/vPQxKKicT7o
          • 7 min
          • Unattended Agents Need Brakes
            A hard spending cap can stop a runaway AI agent, and it can also take your checkout page down with it. This episode covers Cloudflare's spend caps, Anthropic's guide to always-on agents, and a Brake Sheet you fill out before an unattended system goes live. Write down what stops it, who can restart it at two in the morning, and which actions must wait for a person.
            7 min
          • When the Job List Depends on Email
            A two-job phone test for small business dispatch. The office manager supplies approved or made-up job rows. The lead tech leaves Outlook closed and checks whether the job list still opens without going through email. The owner decides whether the current setup works or a controlled backup is needed. The shop in the episode is an example. No customer result or measured benefit is claimed.
            7 min
          • Your AI Agent Can Spend Money

            OpenAI's ChatGPT Wallet lets an agent complete a purchase on your behalf, with per-purchase approval and a network token in place of the card number. John Barrow explains what changed, how narrow the rollout is, what Sierra's Poppy protocol adds, and shares a five-line agent payment gate for CIOs.

            6 min

          About Strata Mind Labs Brief

          From the publisher's feed

          Enterprise AI moves, translated for leaders. A short daily brief on what changed in AI and what it means for the business.