Tech Pulse Daily: June 03, 2026
Curated by Dillip Chowdary at 08:00 IST
Today's Top Highlights
- 01OpenAI Codex: role-specific plugins, in-place annotations, and shareable Sites push Codex beyond pure coding workflows.
- 02GitHub Copilot: the CLI adds /every, /after, local voice input, and a rubber duck critic agent.
- 03Microsoft Foundry: Build 2026 adds production-agent runtime pieces across tools, memory, observability, and governance.
- 04Edge AI stack: NVIDIA JetPack 7.2 brings CUDA 13, MIG, and NemoClaw support to Jetson workflows.
- 05AI security: Anthropic Project Glasswing expands to about 150 more organizations after partners reported 10,000+ high-risk flaws.
This Week in Tech
Microsoft Build sessions cover Foundry IQ, Agent 365, and secure production agents.
Teams should rerun scanners after Kubernetes CVE record propagation.
Pilot windows open for Copilot app, cloud sandboxes, and JetPack 7.2 upgrades.
OpenAI: Codex Becomes a Role and Workflow Surface
OpenAI says Codex now has role-specific plugins, annotations, and a preview for shareable interactive sites, widening the product from engineering automation into broader team workflows.
- Usage signal: OpenAI reports more than 5 million weekly Codex users, with non-developers at about 20% of usage.
- Workflow change: Plugins connect role context, while annotations let users refine generated work in place.
- Builder impact: Sites turns Codex outputs into shareable apps, raising governance questions around data, access, and review.
- Internal deep dive: OpenAI Codex role plugins analysis →
GitHub: Copilot CLI Adds Scheduling, Voice, and Rubber Duck
GitHub Copilot CLI is moving from one-shot terminal assistance toward persistent agent workflows with scheduled prompts, local dictation, and a critic agent.
- Scheduling: /every runs repeated prompts and /after schedules one-off prompts inside a CLI session.
- Local voice: The update records dictation on-device, keeping audio local while lowering the friction for terminal prompting.
- Review loop: Rubber duck critiques plans, implementations, and tests before the main agent continues.
- Internal deep dive: Copilot CLI scheduling and voice analysis →
Microsoft: Foundry Matures Production Agent Controls
Microsoft Foundry used its Build 2026 recap to frame production agents as governed systems with runtime, tools, memory, grounding, models, observability, and policy controls.
- Developer surface: Microsoft Agent Framework adds orchestration blocks, while Foundry Toolkit for VS Code is generally available.
- Runtime path: Hosted agents are expected to reach general availability by early July 2026 with sandboxed sessions, state, and filesystem access.
- Channels: Toolboxes are in public preview, Voice Live adds real-time voice, and Teams/Microsoft 365 Copilot publishing is planned for June.
- Internal deep dive: Microsoft Build agent platform analysis →
Security: ChatGPT Active Sessions Tightens Account Hygiene
OpenAI added Active sessions to ChatGPT settings, giving users a first-party place to review and revoke sessions tied to ChatGPT, Codex, and API Platform where available.
- Control surface: Users can review device, app, approximate location, sign-in time, trusted-device status, and current-session status.
- Revocation: The feature supports signing out of individual sessions or all listed first-party sessions.
- Boundary: It does not manage third-party app sessions, connected apps, Sign in with ChatGPT third-party sessions, or Codex CLI sessions.
- Internal deep dive: ChatGPT active sessions security analysis →
Kubernetes: Unfixed CVE Records Get Corrected
The Kubernetes Security Response Committee is correcting older CVE records whose metadata implied fixes existed for issues that remain architectural risks.
- Affected records: The project highlighted CVE-2020-8561, CVE-2020-8562, and CVE-2021-25740.
- Scanner impact: Corrected metadata can cause vulnerability scanners to surface findings that were previously hidden or misclassified.
- Ops action: Administrators should harden log verbosity, DNS consistency, and Endpoints/EndpointSlice RBAC rather than wait for patches.
- Internal deep dive: Kubernetes CVE cleanup analysis →
NVIDIA: JetPack 7.2 Unifies Jetson Agentic Edge AI
NVIDIA JetPack 7.2 extends the newer Ubuntu 24.04, kernel 6.8, and CUDA Toolkit 13.0 foundation to Jetson Orin while adding agentic deployment features.
- Agent setup: One-command NemoClaw deployment targets secure robotics, industrial automation, and edge AI applications.
- Isolation: Multi-Instance GPU support on Jetson Thor enables predictable multi-workload execution.
- Orin upgrade: Super Mode support on Jetson AGX Orin 32GB gives developers more headroom for physical AI workloads.
- Internal deep dive: JetPack 7.2 edge AI analysis →
Anthropic: Project Glasswing Expands AI Vulnerability Discovery
Anthropic is expanding Project Glasswing, its controlled program for using Claude Mythos Preview to find vulnerabilities in high-impact software.
- Scale-up: The program moves beyond roughly 50 initial partners to approximately 150 additional organizations.
- Finding volume: Initial partners have reported more than 10,000 high or critical-severity flaws.
- Critical sectors: The new cohort spans more than 15 countries and includes power, water, healthcare, communications, and hardware.
- Internal deep dive: Project Glasswing expansion analysis →
Developer Resources
Key Takeaways
Agent tools are becoming environments: Codex Sites, Copilot app sessions, and sandboxes need ownership, retention, and review policy.
Terminal agents need CI-grade controls: Scheduled prompts should be logged, scoped, and blocked from destructive actions without approval.
Session review is now AI account hygiene: ChatGPT Active sessions belongs in offboarding, travel, and incident-response checklists.
Scanner deltas may spike: Kubernetes CVE record corrections can create new findings without a new exploit or patch.
AI security programs are scaling: Glasswing-style rollouts need scoped access, reproducible evidence, disclosure workflow, and patch validation.