Semantic policy execution at API gateways for large language models
Working paper: architectural trade-offs when enforcing probabilistic classifications synchronously along OpenAI-compatible API paths—with constraints on latency, consistency, auditing, and fail-operational ergonomics.
Samuel Oyan — Founder, CEO & Principal Engineer, Intertrace
A runtime-oriented threat taxonomy for agentic LLM systems
Working paper: dimensions for classifying adversarial behaviors that emerge only across multi-step autonomy, nondeterministic tool routing, and cross-tenant integrations.
Samuel Oyan — Founder, CEO & Principal Engineer, Intertrace
Building a simulation lab your red team will actually use
Labs die from friction: stale corpora, no metrics, flaky environments. Here’s how to wire repeatable scenarios, leaderboard signals, and dev-friendly iteration loops.
Samuel Oyan — Founder, CEO & Principal Engineer, Intertrace
Mapping controls to OWASP LLM Top 10 and NIST AI RMF
Translate familiar security frameworks into deployable checkpoints for gateways, evaluations, governance workflows, and evidence exports.
Samuel Oyan — Founder, CEO & Principal Engineer, Intertrace
Runtime verification for autonomous agents
Agents loop, retry, fork subtasks, and call tools asynchronously. Verification must ride alongside execution—not as quarterly manual red-team theater.
Samuel Oyan — Founder, CEO & Principal Engineer, Intertrace
Designing gateway policies that don’t break UX
Blocking is easy; shaping safe assistant behavior users still trust is hard. Patterns for tiers, degrade modes, copy, and rollout that keep latency and satisfaction inside guardrails.
Samuel Oyan — Founder, CEO & Principal Engineer, Intertrace
Prompt injection and indirect attacks in production
Why production LLM workloads face a different class of attacker than labs—and how indirect channels, embeddings, and tool chains turn a single misplaced trust boundary into outage or data exposure.
Samuel Oyan — Founder, CEO & Principal Engineer, Intertrace
Need something else?
Email support@intertrace.ai with topic requests, or try the Simulation Lab and the documentation hub.