Ash Clements
AI Red Teamer · LLM & Agent Security · Threat Hunting
I red-team frontier language models and AI agents, and I build the defenses that assume the attack already won. On Gray Swan's arena I hold a Winner's Circle profile: #26 of 282 in Breaking Ties, 183 breaks in the Hazard Hunt cyber track, and 40 externally judged breaks across 47 frontier models. Behind the rankings is a 123-record corpus that turns every attempt, win or loss, into a falsifiable rule about when a behavior class is breakable at all.
I build the defenses too. AEGIS, a RAG pipeline built on the premise that prompt injection succeeds, tokenizes personal data before it ever reaches the model, so a successful injection walks away with tokens instead of people. Against a plaintext baseline that re-identified 100% of records from real embeddings, AEGIS re-identified 0% while matching retrieval utility, and placed second in Protegrity's 2026 AI Pipeline Security Challenge.
Under it is seventeen years in enterprise security, nine of them at Palo Alto Networks, where I lead Prisma Access and Zero Trust implementations for Fortune 500 organizations across critical infrastructure, own the identity layer, validate public exploits and their variants almost daily, and sit in customer breach and zero-day response. PCNSE certified.
The conclusion I keep landing on is the one I reached about networks a decade ago. No single component can be its own last line of defense, because its judgment can always be talked around by someone patient enough. Intent can be talked around. Actions can be gated.
Experience
SASE & AI Security · Enterprise Architecture · Professional Services
- ›Lead complex Prisma Access (SASE) implementations for Fortune 500 organizations across oil & gas, telecom, and global financial services
- ›Architect Zero Trust network security for high-stakes enterprise environments: critical infrastructure, carrier-grade, global finance
- ›Ran structured red-team exercises on a personal Prisma AIRS lab tenant (prompt injection, tool-use abuse, agent behaviors), and built SCMReady, a read-only parity checker for Panorama to Strata Cloud Manager migrations that was adopted internally, presented to principal and staff engineers, and requested by a Palo Alto Networks Unit 42 engineer for a customer migration
- ›Red-team production language models in Gray Swan's frontier-lab-funded arena, and built AEGIS, a zero-exposure RAG pipeline that placed second in Protegrity's 2026 AI Pipeline Security Challenge
Prisma Access escalations · Lab reproduction · Knowledge base
- ›Owned the hardest Prisma Access escalations, reproduced them in the lab and handed engineering the repro; several became shipped bug fixes
- ›Wrote the internal KBs and customer-facing troubleshooting guides that came out of those escalations
- ›Promoted to Sr. Professional Services Consultant in 2021
Enterprise firewall policy · Network segmentation · PCI-DSS
- ›Migrated enterprise data center firewalls from legacy Cisco ASA to Palo Alto NGFW with automated rule conversion and validation
- ›Operated in a highly regulated PCI-DSS environment with direct exposure to financial-grade security standards
- ›Ran the HA clusters under continuous compliance in production and non-production
Enterprise renewal portfolio · Network-infrastructure contracts · Technical escalation
- ›Managed an enterprise renewal portfolio and served as senior technical escalation point for complex network-infrastructure contracts across the DMV region
- ›Leveraged a deep network engineering background to bridge technical and commercial conversations
Managed service provider · Network & security infrastructure · Team build-out
- ›Delivered network and security infrastructure for enterprise clients: firewall and VPN deployment, device hardening, vulnerability assessment, and remediation
- ›Tasked by the owner to build out a specialized consulting team: hired and mentored engineers, including one who went on to earn CISSP and CISM and serve as an interim CISO
- ›Deepened the enterprise consulting skillset that feeds directly into current Palo Alto Networks work
Core Competencies
Security Tooling
Production tools built at the intersection of AI security, cloud, and enterprise network security.
Second place, Protegrity 2026 AI Pipeline Security Challenge. Personal data is tokenized on ingest and stays tokenized through embedding, the vector store and inference, so a successful prompt injection walks away with tokens instead of personal data. Plaintext re-identified 100 percent of records on real embeddings; AEGIS re-identified none.
Multi-agent LLM security framework. Cross-service attack detection and runtime monitoring, built to pressure-test the runtime threat models AI security platforms have to defend against.
Automated scanner testing AI endpoints against the full OWASP LLM Top 10. Semantic detection catches paraphrased attacks that bypass pattern matching. 35+ payloads, production-ready.
Multi-cloud misconfiguration scanner mapping findings to CIS Benchmarks with executive PDF reporting. Covers AWS, Azure, and GCP in a single read-only scan.
The 2026 conference circuit
Worked the AI security track across Black Hat, DEF CON, and the AI Risk Summit in August 2026, self-funded. If you are building or breaking AI systems and we crossed paths at any of these, the contact form below reaches me.
What I Care About
The work funds the mission. The mission is bigger than the work.
I grew up going to St. George's Camp because people at my church donated so kids like me could go. Now that I do well, I pay it forward, funding spots for underprivileged kids to have the same experience I did.
Shrine Mont / St. George's Camp ↗One of the largest food banks in the United States. I give time and money here because food insecurity is a solvable problem and no one in Phoenix should go hungry.
St. Mary's Food Bank ↗I spend a lot of time in dark corners of the internet. If you see something (child exploitation, animal cruelty), say something. All of these resources accept anonymous reports. You don't have to give your name. You just have to make the call.
Report online child sexual exploitation, anonymous submissions accepted
Report child exploitation and abuse to federal law enforcement
24/7 anonymous crisis intervention and reporting, call or text
International anonymous reporting: CSAM, 50+ countries
How and where to report animal abuse, anonymous tips accepted
Report animal cruelty anonymously, national resources and local referrals
Report animal cruelty in Arizona, anonymous tips accepted
Report animal cruelty anonymously, no name required, potential reward
Contact
Available for consulting engagements, speaking opportunities, and technical advisory roles in enterprise security and AI security architecture.