San Francisco
Joined December 2018
- Tweets721
- Following205
- Followers1.9K
- Likes273
Stand SP04 is open. Day 1 at @GISECGlobal, Hall 1.
Live demo of the labs, stickers to take away, and a straight answer on certifications and team training. Here until Thursday.
#GISEC2026
Day 1 at @GISECGlobal, and the doors are open. We are at Stand SP04, Hall 1, until 17:00.
Labs are running live. Ten certifications mapped to real roles, for one engineer or five hundred.
Dubai Exhibition Centre, Expo City.
#GISEC2026
10 certifications at @GISECGlobal, & 2 of them did not exist 18 months ago.
CAISP & CMCPSE exist because agents & #MCP servers went into production faster than anyone wrote a #threatmodel for them.
Stand SP04, Hall 1.
exhibitors.gisec.ae/gisec-gl…
#GISEC2026 #AISecurity
Three days in Dubai, and the three people you would normally email are standing in one place.
@secfigo, @RajDevSecOps, and @gmaran23 are at the stand at @GISECGlobal for all of it. Labs are running live.
Stand SP04, Hall 1. 16-18 Sep.
exhibitors.gisec.ae/gisec-gl…
#GISEC2026
Your procurement team files a vendor's SBOM as a PDF. That satisfies a checkbox. It does not answer the only question that matters during an incident: are we exposed, and can you prove it in under an hour.
The gap is rarely tooling. Syft and Trivy generate all three formats for free. The gap is a decision nobody's made: which format serves security, which serves legal, and whether you maintain both without duplicating effort.
Detail on the three formats and when to standardize on which: practical-devsecops.com/type…
In mid-2025, a Cursor agent connected to a Supabase MCP server processed a support ticket. The ticket contained embedded SQL instructions. The agent ran them. Tokens got exfiltrated.
Indirect prompt injection is the quiet one. The malicious instruction doesn't come from the user's prompt. It comes from a webpage, a cached file, or a ticket the agent reads as "just data."
None of this requires the model to be dumb. It requires the system around it to trust tool outputs by default, which is how almost every MCP integration works right now. More on these patterns here: practical-devsecops.com/mcp-…
OWASP published its LLM Top 10 in 2023. Most AppSec teams in 2026 are still testing AI systems with the same tools they use for REST APIs.
Most candidates applying have zero hands-on experience with prompt injection testing or tools like Garak and PyRIT, the frameworks these job postings actually require.
Certified AI Security Professional (CAISP) is built around that same OWASP/MITRE ATLAS material, tested through adversarial exercises rather than multiple-choice recall. More detail here: practical-devsecops.com/cert…
