Skip to main content

Frontier AI security research

The lab behind runtime agent security.

Sondera studies how autonomous agents fail, how human intent becomes enforceable policy, and how runtime evidence proves what happened after an agent acts.

Research presented at

Conferences this research has been presented at

  • ICML
  • FLoC 2026
  • Black Hat Arsenal
  • AI Security Forum
  • Unprompted

Lab index

Research that turns intent into a decision you can inspect.

  • ICML AIWILD 2026 Paper

    Autoformalization of Agent Instructions into Policy-as-Code

    A pipeline for translating natural-language agent instructions, tool schemas, and policy documents into checked policy before an agent acts.

  • UnPrompted Talk

    Hooking Coding Agents with the Cedar Policy Language

    A technical talk on bringing policy into the coding-agent loop so agent actions can be constrained before they complete.

  • LLM Solve Paper

    Autoformalization of Agent Instructions into Policy-as-Code

    The same autoformalization research direction as the ICML AIWILD work: turning agent instructions and requirements into checked policy.

Research architecture

From language to a checked decision.

  1. 01 · Grounding

    Tools become schema.

  2. 02 · Model layer

    Intent becomes policy.

  3. 03 · Safety layer

    Policy gets checked.

Lab agenda

Four research problems. One goal: agents you can prove are acting within bounds.

  1. Autoformalization

    Intent becomes policy.

    Instructions written naturally, tool definitions, and requirements are transformed into policy that can be checked before an action runs.

  2. Runtime enforcement

    Actions meet boundaries.

    The security decision moves to the moment an agent asks to read, write, call a tool, or move data across a boundary.

  3. Trajectory evidence

    Behavior becomes inspectable.

    The research program treats agent behavior as a trajectory: context, tool calls, policy attributes, verdicts, and the rule path behind them.

  4. Scalable oversight

    Humans should not review every action.

    Sondera studies how policy, monitors, and evidence can help security teams supervise agent fleets without turning autonomy back into manual approval.

Put the research to work

Prove control before your agents earn more autonomy.