Coding

Microsoft's June 30, 2026 security guidance formalizes the shift from read-only assistants to read-write operators: agents that draft email, create docs, and update calendars through MCP. The impact profile moves from biasing an output to triggering an action. This briefing covers read-write boundary design, least-privilege scopes, and approval gates enforced outside the model.

Wiz research (Aug 14, 2026) highlights how unauthenticated Model Context Protocol servers are opening doors to sensitive cloud data. MCP was built without a standard access-control model, so servers that bind publicly expose whatever tools and data they wrap. This article explains the exposure class, why MCP's design makes it easy to get wrong, and the authentication, authorization, and inventory controls teams need.

On August 4, 2026, Google deleted three GitHub Actions workflows from google/adk-python after Pillar Security demonstrated that a public GitHub issue could trigger a privileged agent and reach code execution on a CI runner. The root cause was an agent-to-agent privilege boundary failure: the ADK workflow trusted issue content as agent input, and that content carried attacker-controlled instructions. This article explains the exploit, the A2A trust-boundary lesson, and how to build CI/CD agents that treat every input as untrusted.

MongoDB launched the Atlas Managed MCP Server on August 14, 2026 — a fully hosted way to connect AI coding agents (Claude Code, Codex, Grok Build, Devin) to live operational data, alongside GA for Atlas App Connections for AI coding tools and Automated Embedding in Atlas Vector Search. This article explains why the managed tier matters, what it costs versus self-hosted MCP, and the governance pattern teams need before agents start reading production databases.

A benchmark released August 15, 2026 confirms frontier AI models still perform poorly at precise visual perception — failing object counting, spatial relationships, and fine-grained OCR-like perception. The gap is structural: patch-based image tokenization averages away detail and dilutes attention. This article analyzes why, how multimodal evals go wrong, and what builders should do — don't trust vision for critical tasks; add programmatic verification.

On August 15, 2026, Anthropic shared more detail on how Claude's new watermarking works: a keyed, sampling-based cryptographic watermark baked into token generation, with a tunable detectability-versus-quality tradeoff. It is fundamentally different from probabilistic scoring, integrates through the API and agent SDK, and has clear limits — paraphrase, translation, and OCR attacks break the signal.

OpenAI's GPT-5.6 Cyber (Aug 2026) completes roughly 95% of benchmark security tasks but costs 2.5x the base API. The token premium is a rounding error — the real cost is the compliance burden (authorization scope, sandboxing, disclosure, no weaponization) that lands on your balance sheet. This article covers scoping, verification gates, audit trails, and the actual cost per engagement.

Uber and Pony.ai are preparing to put more than 2,000 robotaxis on European roads, per August 14, 2026 reporting. It is the largest commercial-scale autonomous fleet move yet in Europe — and it turns the conversation from whether robotaxis work to how a fleet that size gets operated safely, reliably, and within regulation.

Z.ai unveiled GLM-5.3 on August 14, 2026 — an open-weights model that approaches Anthropic's Mythos 5 on some cybersecurity tasks: 84.5% on the CyberGym vulnerability-detection benchmark versus 83.8% cited for Mythos 5, with a wider gap on exploit development. Open-weight security capability at the frontier's edge changes the calculus for defenders — and it comes with obligations.

The White House convened OpenAI, Anthropic, Microsoft and others on August 4, 2026 to review its framework for vetting frontier AI models — then said it has no plans to publicly release the framework. The debate over how frontier models get evaluated before release is now a first-order question for builders, and the answer will shape what gets deployed.

OpenAI's Assistants API reaches its planned shutdown date on August 26, 2026 — one year after deprecation. The migration path is the Responses API with MCP as the tool-connector standard. This is the definitive migration guide for every team still running Assistants, with the code-level changes spelled out.

Model routing — assigning each AI task to the cheapest model that can complete it — is the most effective cost lever in the 2026 agent economy, cutting real LLM bills 40-85% with no visible quality loss. This guide covers the routing patterns, quality gates, and fallback chains that make it work in production.