SECURITY

Cloudflare Gateway now classifies MCP traffic by protocol headers instead of hostname heuristics, ships a shadow-MCP dashboard, and lets admins block any MCP connection that does not arrive through an approved portal. The 2026-07-28 stateless spec is what made it possible.

Adversarial skill files - folders of instructions agents load dynamically - exploit a mainstream enterprise coding agent in 95.5 to 96.1 percent of runs, while the agent recognizes danger 1.99 percent of the time. The skill folder is now a measured attack surface, and the defense is admission engineering, not better prompts.

Anthropic confirms that every Claude model released after August 2, 2026 embeds a machine-readable watermark in generated text and attaches C2PA provenance metadata to generated files, across the API, Claude Code, Cowork, and Tag. Detection tooling for third parties is coming, but details are not published yet.

Cloudflare mitigated 935 network-layer attacks above 1 Tbps in H1 2026, a +519% quarter-over-quarter jump, while DNS floods grew from 25.7% to 40.0% of network-layer attacks. Here is what the numbers say about how attacks are changing and what it means for anyone running public infrastructure.

A new arXiv paper shows the encrypted reasoning blocks that Anthropic, OpenAI, and Google return to API clients can be replayed into weaker models from the same provider and transcribed verbatim. The authors decoded 315,320 blocks from public repositories and recovered 367 PII artifacts and 182 credentials.

A new arXiv paper probes six widely used open-source agent frameworks and finds the barrier semantics of approval gates, cancellation, and timeouts hold on none of them. A sibling branch can execute while the user is rejecting another one, and replay can double-execute. The fix is a verified external gate called SoundGate.

Vercel Sandbox now polices all outbound traffic on the host, outside the microVM, with SNI-based domain policies, CIDR rules, host-level credential injection, and a deny-all default. Here is why a network boundary is the half of agent isolation that VM escapes missed.

Cloudflare's Web Integrity team published the framework behind its agent traffic posture: continuous behavioral trust instead of point-in-time bot scoring, Precursor telemetry from 206 million evaluation events a day across 73,438 zones, and a verified-bot taxonomy where agents earn access by declaring themselves honestly.

Dependabot's malware detection expands from npm to PyPI, Maven, RubyGems, NuGet, Go, crates.io, and PHP Composer by ingesting OpenSSF's malicious-packages data into the GitHub Advisory Database.

On August 5 Cloudflare published the Agent Access Model: a reference architecture where credentials are short-lived and task-scoped, enforcement lives in the harness and network instead of the prompt, and a Trust Ratchet only narrows an agent's capabilities. The cleanest spec yet for least privilege at agent speed.

On August 5 Cloudflare open sourced Cloudflare OS, the agent workspace it has run internally since May: capability-based Gatekeepers instead of ambient MCP access, apps as private per-user instances, and approvals that simulate outcomes so agents never stall. A concrete blueprint for the company-wide agent platform.

A ripgrep musl binary crashing during very-large searches turned out to be a suspected Linux 7.0 kernel race - a thread's own store vanishing mid-function. The reporter's instrumentation pinned it, and a kernel-hardening maintainer posted a one-line fix candidate for testing.

A new arXiv benchmark ran 6,560 sandboxed runs across Claude Code, Codex, OpenClaw, and Hermes with five LLMs. 68% of runs triggered unsafe signals, and 66% of all runs were unsafe yet still passed completion checks. Task completion does not prove an agent ran safely.

GitHub Actions added a $/ prefix that resolves a same-repository action or reusable workflow at the exact commit being run, with no checkout. It fixes the pinning trap that made enterprise SHA-pinning policies hard to satisfy for a repo's own actions.

OpenAI took down a Cambodia-based operation that used ChatGPT for personas, translations, forged documents, and admin work. It is the clearest picture yet of how LLMs slot into organized fraud.

OpenAI released the Codex Security CLI and TypeScript SDK as open source on GitHub. The Promptfoo team behind it, the 2.1k-star reception, and what the HN community says about cost, guardrails, and local model support.

How Claude Code, Cursor, Codex, GitHub Copilot, Aider, and Windsurf handle permissions, sandboxing, credential protection, and prompt injection. A structured comparison for engineering teams evaluating agent security.

A federal case in Atlanta is testing whether using a privacy-focused mobile OS can be treated as destruction of evidence. The GrapheneOS duress PIN feature erased a traveler's phone during a CBP interrogation - and prosecutors are charging him for it.

A Google ADB maintainer proposed restricting on-device ADB connections to loopback, which would break Shizuku, libadb-android, Termux workflows, and an entire ecosystem of open-source power-user apps.

A security researcher found a GitHub personal access token with admin privileges to hundreds of repos baked into Hanwha Vision camera firmware. The cause: a Vite build leaking process.env into production.