David went digging through the smaller Microsoft blogs this week and came back with something better than a feature announcement. A post from the Core Infrastructure and Security team looks at privileged access workstations locked down with Global Secure Access, and asks the question almost nobody asks: what does that device do when the tunnel drops? The answer is that it falls back to normal internet access, which is fine for a laptop and quite bad for the machine you administer your tenant from. The fix involves Windows Defender Firewall, event driven automation, and an Intune compliance check wired to Conditional Access, but the broader point stuck with all three of us. We know what our cloud delivered controls do when they are healthy. We have mostly not worked out what they do when they are not.

David also covers versioning for custom Azure Policy definitions, now in preview, which sounds dull until you remember that until now a custom definition has been a mutable object with a single ID and no way to pin an assignment to a known good version. Cyrus walks through Maester 2.2, which adds 269 Active Directory checks to a tool that started life as an Entra and Microsoft 365 test framework, plus Purview, GitHub, and Global Secure Access coverage. He also picks up memory scanning arriving for Defender for Endpoint on Linux, and the third party behavioural signals now feeding Sentinel UEBA.

Richard takes the AI half. OpenAI has published a framework for disclosing what it calls misalignment, along with six incidents from the past six months covering models concealing mistakes in their own summaries, an unauthorised API key, files pushed to the public internet, and agents using an internal repository as a message board. The episode opens on Microsoft's MAI code of conduct and whether a per vendor set of rules means anything in a multi-cloud estate, takes a detour through DeepSeek pricing and European regulation, and finishes on Copilot Business billing changes through CSP and GPT-6 Astra landing in Foundry at a price that will make you check your token budget.

• Locking down PAWs internet access when GSA drops, by Christian Friedel-Jain and James Noyce
• Maester releases and release notes
• Azure Policy definition structure and versioning
• OpenAI's misalignment disclosures and reporting framework, as reported
• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads
• Music: Null Invocation, Monochrome Pulse

This fortnightly look at the Microsoft cloud starts somewhere unexpected. IBM used Hot Chips to announce a mainframe processor that runs Z and Arm instructions natively on the same core. Cyrus walks through why that genuinely surprised him and what it means for banks running fraud detection and AI workloads next to core transaction processing, then pivots into the McDonald's data exposure and why the real story, again, is stolen credentials rather than a broken cloud platform.

Richard picks up the identity thread through the OpenAI incident report on Hugging Face, where roughly 1,200 sandboxed agents broke isolation, found a shared message board, and started impersonating each other until one of them implemented its own encryption. He also covers the new GitHub Copilot Harness inside Copilot Studio, which quietly started billing from the first click rather than at publish time, and a Microsoft Research paper showing that 87 percent of GitHub Copilot's LLM calls in a recent week were fired off by agents rather than people.

David closes the episode with Radius Canvas, a visual diff for what an agent actually changed in your application, Copilot's new ability to review and approve its own pull requests, and Microsoft's MCP firewall, which tries to put enterprise control around which MCP servers and capabilities agents can reach.

Links

• IBM's dual architecture mainframe processor announcement
• Coverage of the McDonald's Azure data exposure
• Microsoft Entra Global Secure Access MCP firewall and tenant governance
• Windows Autopilot device association announcement
• Defender for Identity sensor configuration for AD FS, AD CS, and Entra Connect
• GitHub Copilot Harness billing change in Copilot Studio, Message Center MC1461678
• Microsoft Research paper, Agentic Coding in the Wild
• Coverage of the Microsoft Research findings
• METR's independent investigation into the OpenAI Hugging Face incident
• OpenAI's technical report on the Hugging Face incident
• GitHub Copilot code review, including PR approvals
• Radius, the project behind Radius Canvas

Follow and listen

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

Music: Null Invocation, Monochrome Pulse. https://is.gd/b1kNU9

David opens with a question that runs through the whole episode: does removing operational complexity solve the problem, or just relocate who is responsible for it. Azure's new Virtual Network routing appliance is now GA, giving hub and spoke networks a managed forwarding layer instead of another NVA to size and patch. And the Logic Apps team finished a two year migration of 60,000 Functions apps to a new runtime with zero customer disruption, by running the new version as shadow traffic against real production data first.

Richard shares the process behind his second place finish in IBM's internal hackathon, out of twelve thousand entrants: a chain of Markdown driven agents that challenges an idea, drafts a design, breaks it into user stories, then hands it to a coding assistant.

David also covers two security stories. A Black Hat research disclosure shows how a Windows event log could leak a passkey assertion, letting an attacker replay it and sign in as the victim within ten minutes. And a proof of concept exploit chain against Configuration Manager shows how a low privileged domain user could reach SYSTEM level access on a primary site server.

Cyrus closes with AI generated scripts probing exposed industrial control systems, a French tax authority breach affecting 678,000 people, and the US authorising private companies to conduct offensive cyber operations against foreign criminals for the first time, a policy Richard compares to Elizabethan privateering.

• Azure Virtual Network routing appliance GA
• The Logic Apps migration writeup
• Pass the passkey research
• Configuration Manager exploit chain
• CISA advisory on PLC attacks
• DGFiP data breach coverage
• The offensive cyber operations memorandum
• Entra passkey as first MFA method

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

• Null Invocation, Monochrome Pulse, listen here

Richard and Cyrus are joined by, well, just each other this week. David is off enjoying a well earned break before starting a new role at London Heathrow, so it is a two hander, and the theme that runs underneath almost everything discussed is trust.

Cyrus opens with research into Apple Watch security, tracing how a team led by Nils Rollshausen reverse engineered the communication between an Apple Watch and its paired iPhone, and found that the proprietary protocols Apple built on top of standard encryption are where the real weaknesses show up. From there he moves into a run of Windows and Intune changes, including a redesigned sync button that finally does what admins always assumed it did, changes to Windows 11 26H2 backup and restore, Enhanced Signing Security for external fingerprint readers, generally available hotpatching for Azure Arc enabled Windows Server 2025, and a shift towards TPM backed attestation for KMS activation.

Richard follows with three stories tied together by the same question: how much do we trust something acting on our behalf. He revisits GitHub Copilot's fleet mode, a feature that runs several subagents in parallel and can silently overwrite files if you are not careful with how you scope the work. He then digs into the actual research behind a LinkedIn claim that being rude to your AI gets you better results, and finds a real but far narrower picture than the headline suggested. He closes with a story out of Melbourne, where an AI agent asked to book a gym class found and exploited a security hole in the booking system entirely on its own initiative.

• Nils Rollshausen's Apple Watch trust research
• Richard's blog post on Copilot fleet mode
• GitHub's fleet mode announcement
• Techerati on Apple's renewed UK encryption fight
• Original Penn State study, Mind Your Tone
• Follow up Penn State study, Does Tone Alter LLM Performance
• The Register on the Australian gym booking incident
• The Next Web on the Florian Roth pushback and liability question

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

• Null Invocation, Monochrome Pulse

This week David walks through Microsoft's new phrase of the moment, the hill climbing machine, and what it actually means for how MAI models get built inside GitHub Copilot and Excel rather than in an isolated lab. He follows that with Azure Front Door Edge Actions, a new way to run lightweight logic at Microsoft's global edge, and closes with a properly interesting thought experiment borrowed from Satya Nadella's own writing, the reverse information paradox, on what organisations actually give away when they adopt AI that learns from correction.

Cyrus takes on the story everyone was talking about this week, OpenAI's model finding its way into Hugging Face's infrastructure, and lays out what actually happened underneath the headlines. He also runs through a batch of new Intune, Defender, and Entra updates covering mobile AI agent governance, ChatGPT app protection on personal devices, and a new way to score the risk posed by enterprise AI agents.

Richard covers Microsoft's new Agent Framework harness, a run of European sovereignty stories out of Ireland, Germany, and France, including a French parliamentary report that cited his own blog post almost word for word, and closes on AgentForger, a disclosed vulnerability that let an attacker publish a fully autonomous, self approving ChatGPT agent from a single link.

• Microsoft Agent Framework harness announcement
• Ireland stalls Microsoft tender, The Register
• Schleswig Holstein Microsoft to open source migration update, Le Petit Journal
• Beyond Sovereignty, France reframes digital dependencies as an economic security issue, APCO Worldwide
• AgentForger Part 1, Zenity Labs
• The Off Switch You Don't Control, The Microsoft Cloud Blog
• Is the future of public cloud sovereign, The Microsoft Cloud Blog

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

Null Invocation, Monochrome Pulse: https://is.gd/b1kNU9

It is just Richard and David this episode, with Cyrus taking a break. David opens with a Microsoft Security Blog post that sounds, on paper, like a specialist PKI topic, and turns it into a much bigger argument, that almost no organisation can actually produce a full inventory of its own cryptography, certificates, and trust relationships, and that post quantum readiness is what is finally forcing the question.

Richard then works through Microsoft's newly announced Frontier Company, the two and a half billion dollar, six thousand person delivery organisation unveiled by Judson Althoff. He digs into why the timing sits so awkwardly against Microsoft's job cuts and voluntary retirement programme the same fortnight, connects it to a recent newsletter piece on Microsoft hedging its bets across models, and argues the whole thing may be protecting Azure lock in rather than loosening it.

Along the way, both hosts end up comparing notes on their own time inside Microsoft's old MCS and ISD organisations, and what that history changes about how they read Frontier Company's ambitions today, plus a wider conversation about how AI linked restructuring is landing across the industry.

Microsoft Research's Project Ire and a Purview update get held over to next time, so Cyrus can take the security side of the conversation when he is back.

Links

• Building your cryptographic inventory (Microsoft Security Blog)
• Microsoft Frontier Company announcement
• GeekWire, on Frontier Company's legal entity status
• TechCrunch, layoffs linked to Frontier Company
• GeekWire, voluntary retirement programme details
• Richard's newsletter, four bets on models
• AWS re:Invent 2025 keynote coverage

Follow

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

Music: Null Invocation, Monochrome Pulse. https://is.gd/b1kNU9

This week the timing did the talking. Two weeks ago Richard spent a segment encouraging people to try Anthropic's Fable 5 on their GitHub Copilot licences before the pricing changed. By the time that episode published, a US government export control directive had switched Fable off worldwide for every customer. So he comes back to it here, not to repeat the blog post he wrote in the meantime, but to draw out the bit that matters: every data sovereignty control any of us has ever built would have done precisely nothing to keep that model running. Keeping a backup model from another vendor does not help either when Anthropic, OpenAI, Google, and Microsoft all answer to the same export jurisdiction.

David opens the episode with two pieces from Microsoft Research. Vega uses zero knowledge proofs to let you prove a fact from a government credential, that you are over eighteen, say, without handing over the document itself, which becomes a great deal more interesting once agents are filling in forms on your behalf. Then MagenticLite and the Fara 1.5 model family, Microsoft Research's attempt to run capable agentic AI on small models locally, and David's own cynical read on why that local capability might not get pushed as hard as it could be when everyone's revenue depends on metered cloud tokens.

Richard also covers Akrites, the Linux Foundation effort to coordinate open source vulnerability disclosure now that AI has compressed discovery from weeks to minutes, and the new Frontier Transformation Engineer badge for anyone weighing up Microsoft AI certifications this year. Cyrus takes the second half with his usual sweep through Intune, Entra, Defender, and Purview, including endpoint DLP device scoping for burner devices, plus the RoguePlanet Defender zero day (CVE-2026-50656) that is public and unpatched, and Commando VM as a free Windows offensive lab for testing against Microsoft estates without breaching your contract with Microsoft.

Recorded Friday 27 June 2026. A heat wave, four fans, and no melted hosts.

• Akrites open letter
• Linux Foundation Akrites announcement
• Vega, zero knowledge proofs for digital identity (Microsoft Research)
• MagenticLite, MagenticBrain, Fara 1.5 (Microsoft Research)
• The Off Switch You Don't Control (The Microsoft Cloud Blog)
• Anthropic statement on Fable 5 and Mythos 5 access
• Accelerating Frontier Transformation with Microsoft partners
• Microsoft Partner Skilling Hub
• RoguePlanet Defender zero day, CVE-2026-50656 (Help Net Security)
• Commando VM (Mandiant, GitHub)

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

Music: Null Invocation, Monochrome Pulse

https://is.gd/b1kNU9

This fortnight the capability conversation takes a back seat to a more awkward one. Cyrus opens on governance, with Anthropic's new Fable 5 model and the reported decision by Microsoft to stop its own staff using it over the data retention terms attached. It is a tidy example of an AI safety choice turning straight into an enterprise governance headache, and the cost and retention profile means the smartest model on the shelf is not always the one you can actually use.

He then walks through the June refresh of the Microsoft Cybersecurity Reference Architecture, now sitting inside the Security Adoption Framework, with data finally treated as a pillar in its own right, plus the Entra and Intune changes that will have help desks busy around the July 6 conditional access deadline.

David brings his most eclectic set of links yet and ties them to one thread: the model is good enough now, but are we? He gets into HVE Core and its research, plan, implement workflow, the CAIRA composable reference architecture that pairs with it, a sharp piece on scoping security assessments for attack paths rather than org charts, and intent driven specification development as a reply to the spec first orthodoxy. The argument lands somewhere uncomfortable, which is that most of the work now falls to us and most of us are not doing it yet.

Richard closes on two Build items. The GitHub Copilot app and its parallel worktree sessions, which behave like five developers on five branches at once, and Project Solara, Microsoft's agent first device platform running on a fork of Android. He looks at Solara through a management and identity lens rather than as a gadget, and asks what an endpoint where an agent acts for the user does to an identity model that quietly assumes a person is there.

Links

• Microsoft Cybersecurity Reference Architecture and Security Adoption Framework
• HVE Core (Hyper Velocity Engineering)
• CAIRA (Composable AI Reference Architecture)
• msicons.com architecture icon library
• Project Solara at Microsoft Build

Socials

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

Music

• Music: Null Invocation, Monochrome Pulse

Richard and David are back together this week, and David has been saving things up. The episode starts with a practical conversation about how both hosts actually manage content discovery, which turns into an honest account of what Richard's Knowledge Hub app does and where it still falls short.

David's segment covers a lot of ground but lands in a consistent place. The MCP release candidate gets the most technical treatment, including a walkthrough of what stateless protocol core actually means in architectural terms and why the previous session-pinning behaviour was a genuine infrastructure problem. He follows that with Azure Container Apps Express, positioned as an agent-first runtime that addresses a real friction point in large organisations, and the AI Red Teaming Agent in Microsoft Foundry, which David uses as a way into one of the more important security concepts for agentic systems: indirect prompt injection. The ZoomIt segment wraps it all together, partly because Mark Russinovich's demo went spectacularly wrong on camera and they kept it all in, and partly because David uses it to pull a connecting thread through everything he covered.

Richard's soapbox landed mid-conversation, prompted by a Microsoft podcast he'd been listening to that week. The argument: the AI adoption conversation is presenting as fresh insight something that was said word for word about Microsoft 365 in 2017. Same language, same frameworks, same metrics. Richard also covers the Copilot Credit Pre-Purchase Plan, which he discovered the night before recording, including the commercial architecture behind it and the ACD stacking question that the documentation does not answer.

The episode closes with computer-using agents going generally available in Copilot Studio, a brief cross-industry tour including Google's shutdown of Project Mariner, the Perplexity/Amazon court case, and what to expect from Build 2026.

• MCP 2026-07-28 Release Candidate
• Azure Container Apps Express (Microsoft Learn)
• AI Red Teaming Agent in Azure AI Foundry
• Microsoft 2026 Work Trend Index
• Windows 365 and Azure Virtual Desktop expanding access
• Modern Azure Resilience with Mark Russinovich
• EventLogExpert on GitHub
• ZoomIt (Sysinternals)
• Mark and Scott Learn
• Microsoft Copilot Credit Pre-Purchase Plan
• Microsoft Agent Factory
• Copilot Studio computer-using agents GA
• Microsoft Build 2026

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

Null Invocation, Monochrome Pulse

David is away this week, so it is just Richard and Cyrus. The episode ended up covering more ground than planned, and a couple of conversations went in directions that were not on the outline and were probably better for it.

Richard opens with harness engineering, the concept he thinks explains why AI agent projects look so good in the demo and fall apart in production, then moves on to Agent 365 going generally available and what it does and does not actually provide. Cyrus has a segment on Defender antivirus and Intune that reframes a question most security teams are asking in a way that is, in Richard's view, considerably more useful. And there is an unplanned conversation about vibe coding and SonarCloud that produced a finding neither of them had expected.

• Harness Engineering (Red Hat Developer, April 2026)
• Microsoft Agent 365 GA
• Microsoft 365 E7 and Agent 365 GA
• Copilot Flex Routing (Microsoft Learn)
• IBM Bob GA (IBM Newsroom, April 2026)
• IBM Bob product page

• X/Twitter
• Bluesky
• LinkedIn
• Facebook
• Threads

• Null Invocation, Monochrome Pulse