ls writing/ (15 entries)

Writing

  1. Why more agent autonomy creates a security architecture problem

    Draft: work in progress

    Authority, delegation, trust boundaries and least privilege, once the thing acting on your behalf makes its own decisions.

  2. Why agent memory is not just a vector database

    Draft: work in progress

    Task state, provenance, authority and time all matter. Similarity search alone captures none of them.

  3. Agents need a control plane, not just a system prompt

    Draft: work in progress

    A system prompt is a request, not a boundary. Tool-using agents need enforcement that sits outside the model.

  4. Pentesting cloud IAM as a graph problem

    Draft: work in progress

    Roles, trust policies, STS and cross-account access form a graph. Testing it means finding paths, not misconfigurations.

  5. Designing cyber agents that know their rules of engagement

    Draft: work in progress

    Scope should be enforceable machine state, not prose buried in a prompt.

  6. Dynamic delegation: letting agent topology follow the problem

    Draft: work in progress

    Fixed multi-agent org charts may be the wrong abstraction. What if structure formed around each task instead?

  7. Building an evidence-first AI pentesting agent

    Draft: work in progress

    Findings, tool calls and conclusions should all resolve back to evidence. A design for AI-assisted testing that keeps humans in charge.

  8. From pentester to adversarial systems engineer

    Draft: work in progress

    Offensive security increasingly runs through cloud, identity, APIs, CI/CD and autonomous systems. The job is changing shape.

  9. Local AI as security infrastructure

    Draft: work in progress

    Privacy, model control, isolation and reproducibility: why running models yourself matters beyond cost.

  10. Building offensive infrastructure as infrastructure

    Draft: work in progress

    C2, telemetry, test environments and deployment pipelines deserve the same engineering discipline as production systems.

  11. Recursive agents and the context problem

    Draft: work in progress

    How does a subagent know what it needs without inheriting the entire parent context?

  12. Running 27B models on constrained hardware

    Draft: work in progress

    Quantisation, the VRAM/RAM split, context size and throughput: lessons from running mid-size models on hardware that is not built for it.

  13. The approval came after the action

    Draft: work in progress

    An agent changed a file before the approval prompt said it was allowed to. Why UI-level approvals are not enforcement.

  14. TTFT is often more important than tokens per second

    Draft: work in progress

    For interactive and agent workloads, waiting for the first token shapes the experience more than raw generation speed.

  15. What adversary emulation is actually for

    Draft: work in progress

    Not a trophy hunt. Attack-path discovery, telemetry, visibility gaps and proving that detections and playbooks work.