AI Agent Security Guides: Prompt Injection, CVEs, Defenses

The 2026-07-28 spec removes the initialize handshake and Mcp-Session-Id header. A maintainer’s migration guide with before/after request diffs, the new routing headers, the Extensions framework, deprecations, and OAuth 2.1 hardening.

HalluSquatting turns predictable AI hallucinations into malware delivery. Why lockfiles and cooldowns do not stop it, plus the PreToolUse hook, sandbox config, and .npmrc settings that do.

CVE-2026-42271 chains with the Starlette BadHost bypass (CVE-2026-48710) for unauthenticated RCE on the AI gateway. Detect exposure, upgrade to 1.83.7, rotate every key, and harden the MCP test endpoints.

Set up OpenAI Codex Security on GitHub end-to-end: connect a repo, edit the threat model, triage validated findings, ship patches as PRs. 74% TPR vs Semgrep 20% and Snyk 28% in independent testing.

Defend Claude Code workflows against the April 2026 Comment and Control CVE. Tool allowlists, OIDC via Bedrock, script caps, egress blocks, and a before/after hardened workflow.

Workflow YAML, false positive tuning, token cost math, and a layered pipeline with Semgrep and Snyk for Anthropic’s official security review action.

Anthropic locked Claude Mythos to 12 Project Glasswing partners. What the benchmarks, pricing, and restricted access mean for everyday developers, plus what to do while waiting for public release.