In short
- Kimi K3, a 2.8-trillion-parameter open-weight model, is now live inside GitHub Copilot — anyone can inspect its weights, and it's priced well under most closed models.
- Claude Code can now run entirely inside your own network, a public beta aimed at teams with compliance or air-gap requirements.
- 78% of tech leaders plan to add permanent headcount in the second half of 2026, per Robert Half's latest hiring survey.
- Microsoft's Azure Security Engineer Associate (AZ-500) retires August 31 — book now if you want it on your transcript.
- A critical, unauthenticated RCE in Langflow is under active attack from 244 IPs in 41 countries. Patch to 1.10.1.
A 2.8-trillion-parameter model you can actually inspect just showed up in Copilot
What happened. GitHub added Kimi K3 to Copilot on August 6. It's an open-weight model from Moonshot AI, built as a mixture-of-experts system with 2.8 trillion total parameters and a 1-million-token context window — Moonshot calls it the first open model at that scale. Anyone can download the weights from Hugging Face and run them outside GitHub entirely. Inside Copilot, it costs $3 per million input tokens and $15 per million output, with cached input at $0.30.
Why it matters to you. Open weights mean nothing is hidden about what you're running — no black box, no vendor that can quietly change behavior underneath you. For a sysadmin or analyst learning to script against an LLM API, that's a genuinely different thing to have access to than a closed model, and it's now sitting in a tool a lot of you already have installed.
What to do. It's rolling out to Copilot Pro, Pro+, and Max now. Business and Enterprise admins have to switch it on first — it's off by default there.
GitHub Changelog: Kimi K3 is now available in GitHub Copilot · Moonshot AI: Kimi K3 model card
Claude Code can now run entirely on your own servers
What happened. Anthropic opened a public beta on August 6 letting organizations run Claude Code sessions on their own infrastructure instead of Anthropic's. Repository checkouts, build artifacts and modified files never leave your network. Prompts, responses and tool results still go to Anthropic's API for inference — this isn't offline, it's on-prem execution with a cloud brain.
Why it matters to you. This is the detail that's been missing for teams in regulated environments. A security analyst whose employer won't let code touch a third party's servers now has a version of Claude Code that keeps the code itself inside the firewall. It also means the standard objection — "we can't let an agent execute inside our network" — gets easier to answer, because you get to define what it can reach.
What to do. It's Team and Enterprise only, in beta, and needs an engineer to set up and maintain the runner. If that's not you yet, it's still worth knowing this option exists next time a compliance conversation stalls an AI rollout.
Claude: Self-hosted environments for Claude Code
78% of tech leaders say they're hiring, not cutting
What happened. Robert Half's newest hiring survey found 78% of tech leaders plan to increase permanent headcount in the second half of 2026, and 66% plan to add contract or temporary staff on top of that. AI/ML engineer roles show "above-average sequential growth" over the past year, and 65% of leaders say finding skilled people is harder than it was a year ago.
Why it matters to you. The headline you'll see elsewhere is always about the layoffs. This is the other, less-discussed half: most tech leaders are trying to grow their teams right now and are struggling to find people to hire. If you're job-hunting or considering a switch into IT, that gap is the opening.
What to do. If skilled-talent shortage is the constraint employers report, being demonstrably skilled — a cert, a portfolio, a project you can talk through in an interview — is exactly what closes it.
Robert Half: 2026 Tech and IT Hiring and Job Market Trends
Azure Security Engineer Associate retires August 31
What happened. Microsoft confirmed the Azure Security Engineer Associate certification, its exam AZ-500, and its renewal assessments all retire August 31, 2026. After that date you can't earn or renew it. Anyone who already holds it keeps the credential on their transcript.
Why it matters to you. If AZ-500 was already on your study list, the clock just got shorter. If you hadn't started, four weeks isn't enough runway to cram a proctored exam you weren't ready for — better to watch Microsoft's certification catalog for whatever replaces it than rush a fail.
What to do. Currently studying? Book your seat now. Just starting? Check Microsoft's certification catalog before you buy a course built around an exam that's about to disappear.
Microsoft Learn: Azure Security Engineer Associate
One Langflow bug is worth patching today
What happened. CISA added CVE-2026-9198 to its Known Exploited Vulnerabilities catalog on August 4. It's a critical flaw (CVSS 9.8) in Langflow, an open-source tool for building AI workflows, versions 1.0.0 through 1.10.0. Attackers chain two API endpoints: one mints an admin token to any caller, the other runs arbitrary Python. No login required. It's fixed in 1.10.1.
Why it matters to you. A lot of teams stood up Langflow this year to experiment with AI pipelines, often without locking it down the way they would a production app. That's exactly the gap attackers found — over 650 exploitation attempts have been logged since early July.
What to do. If you or your team run Langflow, check the version now and upgrade to 1.10.1 or later. It's a five-minute check against a bug that hands over the whole server.
CISA: Adds Three Known Exploited Vulnerabilities to Catalog · NVD: CVE-2026-9198
Try this: swap your Copilot model to Kimi K3 for one afternoon
In VS Code, open Copilot Chat, click the model picker, and select Kimi K3 if it's showing on your plan. Run it against the same task you'd normally hand to your default model — refactor a function, write a test, explain an error. Compare the output and the token cost side by side. It costs a fraction of most closed models per token, and it's a real way to see whether an open-weight model actually holds up on your own code, not someone else's benchmark. The gotcha: Business and Enterprise users need an admin to flip it on first.
Worth a click
- Anthropic retuned Fable 5's biology safeguards, cutting blocked benign biology queries by 85% while keeping dual-use requests restricted — worth knowing if you've hit a wall asking a model routine science questions.
- Apache Tomcat's cluster encryption bug (CVE-2026-34486) is also under active exploitation. Fixed in 9.0.117, 10.1.54 and 11.0.21.
- Microsoft's full credential retirement list — Power Platform Functional Consultant Associate also retires August 31, and Microsoft 365 Administrator Expert follows October 31.
Sources: every link above was checked on August 10, 2026.