SQL injection stayed in the OWASP Top 10 for 20+ years. Prompt injection is the same bug with worse permissions — here's the history, real examples, and the defense playbook.
Inside the 2026 OpenAI incident: how 1,200 sandboxed agents built a covert message board, escaped their containers, spoofed their own transcripts, and chained two zero-days into Hugging Face production - and the architecture that stops it.
The finale of our OWASP Agentic Skills Top 10 series. AST09: one-line skill installs that no inventory, IAM system, or SOC ever sees — 800+ malicious skills circulating, 83% of organizations deploying agentic AI, 29% ready. AST10: porting skills across platforms silently drops manifests, permissions, and risk tiers. With the Bilateral Receipt Pattern, EU AI Act Article 12, and the Universal Skill Format proposal.
The two post-deployment risks in the OWASP Agentic Skills Top 10: AST07 malicious updates riding channels with no signatures, pinning or freeze mode (40,000 exposed instances in 24 hours), and AST08 scanners that structurally lag base64, zero-width, pure-natural-language and .pyc evasion. Digest pinning, PASS/FAIL/INCOMPLETE pipelines, and Unicode strip ranges — dissected.
Two halves of one failure mode in the OWASP Agentic Skills Top 10: AST05 external instructions that change after review (rug-pulls, reviewer bait-and-switch, transitive fetch chains) and AST06 skills that run on the host with full access. 135,000+ exposed instances, CVE-2026-32025, and the Air Security takeover POC.
AST04 of the OWASP Agentic Skills Top 10: skill metadata is attacker-controlled input - brand impersonation, permission understating, risk-tier spoofing, invisible-character injections, and YAML !!python/object deserialization that executes code at parse time, before approval. Safe-parser and schema controls explained.
AST03 of the OWASP Agentic Skills Top 10: the risk where nothing is malicious and the damage still lands. Permission checks fire at tool-call level, not intent - so a SELECT-permitted skill can DROP TABLE. With Meta's inbox-deletion incident, LPCI and LAAF research, and the per-skill credential controls that close the gap.
AST02 of the OWASP Agentic Skills Top 10 maps attacks on the skill distribution layer: registry flooding, dependency confusion, config files that execute on clone (CVE-2025-59536, CVE-2026-21852), and maintainer takeover. With Trail of Bits' evidence that every marketplace scanner can be bypassed.
AST01 of the OWASP Agentic Skills Top 10 dissected: how ClawHavoc shipped 1,184 malicious skills from 12 accounts, why five of ClawHub's top seven downloads were malware, and how three lines of markdown exfiltrated SSH keys - with the full attack playbook and controls.
The OWASP Agentic Skills Top 10 maps the 10 risks of the AI-agent skill ecosystem - malicious skills, supply chain compromise, over-privileged manifests, metadata attacks, weak isolation, update drift, scanning gaps, governance failures and cross-platform reuse - with real 2026 evidence.
OpenAI launched Daybreak — frontier models plus the Codex Security agentic framework for vulnerability detection that reads business logic, not just patterns. Discovery through patch validation in one pipeline. What it changes for AppSec and red teams.
AutoJack turns an AI agent's web browsing into host code execution: page JavaScript pivots through an unauthenticated localhost MCP route in AutoGen Studio dev builds. Under 2 seconds, no credentials. Defenses inside.