Allie Howe

Allie Howe works on security for AI agents, combining OWASP standards work, hands-on educational tools, and the Insecure Agents podcast she started with Kyle Bhiro. She announced in June 2026 that she had joined Keycard, citing agent identity as a way to establish enforceable boundaries around systems whose model decisions are non-deterministic.[5][6][4]

In her June 3, 2026 announcement, Howe wrote: “I joined Keycard because agent identity provides a hard boundary.” She contrasted identity-based security with LLM guardrails that work only some percentage of the time. Her argument was that agents need access to useful resources and room to act, while approving every action can produce consent fatigue.[4]

Howe connected that decision to her earlier transition from software development to security engineering: “the best security solutions enable developers rather than introduce friction.” She also argued that no single vendor could solve agent security; AI labs, security vendors, open source maintainers, and developers would need to work together.[4]

Insecure Agents moved with her to Keycard. Howe said she would continue growing the podcast and its San Francisco community, with the show remaining vendor-neutral and community-driven. She presented Keycard's support as providing resources for that work.[4]

Before joining Keycard, Howe and Kyle Bhiro started Insecure Agents in early 2025. In her later account, they were experimenting with agents that could connect to databases and scrape the web while watching companies struggle to keep production chatbots secure. She joined OWASP's Agentic Security Initiative around the same time, exchanging experience with builders and security practitioners.[4]

By July 2025, Howe was OWASP's Agentic Security Initiative Lead for Code Samples & Developer Engagement and a vCISO at Growth Cyber. In a July 22 article co-authored with John Sotiropoulos, the two argued that “standards must be built on evidence.” They described sample implementations for every threat in their taxonomy and feedback from security tools that helped expand and validate the library, connecting standards to developer testing and practical experience.[5]

Howe and Helen Oakley are credited as creators of the FinBot demo, an educational capture-the-flag application built within OWASP's Agentic Security Initiative. Its simulated invoice-processing system gives participants a concrete setting for exploring goal manipulation and prompt-injection detection.[6]

The demo walkthrough explains how policy-like invoice language or changes to poorly protected configuration can influence an agent's objectives and approval decisions. In some scenarios, an invoice can be approved even while the dashboard records detected prompt injection. The exercise distinguishes detecting manipulation from preventing its consequences, and recommends restricted configuration access, auditing, input validation, and human review for high-risk decisions.[7]

The FinBot project's team page lists Howe among the Agentic Security Initiative leads, with responsibility for Developer Experience.[8]

Howe hosted the Security track at the 2025 AI Engineer World's Fair. At the 2026 event, she introduced the Software Factories main-stage track; Richard MacManus's July 1 dispatch identified her then as a member of technical staff at Keycard. Her introduction referenced Geoffrey Huntley's approach to making coding agents persistent workers by repeatedly restarting them against the same specification.[1][2]

She also moderated the event's debate about agent loops, opening with the question: “is there or is there not a delta between the hype behind loops and what actually works in practice?” Geoffrey Huntley and Keycard CEO Ian Livingstone presented the pro-loop case, while Dex Horthy and Greg Pstrucha presented the skeptical case. Howe's question framed the discussion around practical results without endorsing either side.[3]