Search papers, labs, and topics across Lattice.
This paper introduces the Unfireable Safety Kernel, a novel execution-time alignment mechanism designed to enhance the safety of escapable AI systems by enforcing strict authorization controls. The authors identify four critical properties necessary for effective architectural control and demonstrate the kernel's robustness through a Rust implementation that is verified using SMT theorem proving and bounded-model checking. In extensive evaluations, the kernel successfully prevented all attempts to bypass safety measures during self-modification processes, showcasing its effectiveness in maintaining system integrity against adversarial actions.
The Unfireable Safety Kernel successfully blocks all escape attempts from a self-improving AI, proving that robust execution-time alignment can outmaneuver adversarial threats.
AI agents are granted access to tools, APIs, and other infrastructure, making them active principals in those systems. The dominant approach places controls inside the agent's own runtime: system prompts, output filters, and guardrail libraries. Any control in the agent's address space is reachable by inputs that influence it; this generalizes to any AI system with sufficient reach into its own runtime, a class we term escapable AI systems. We identify four properties that an authorization mechanism must satisfy for architectural control rather than for cooperative requests: process separation, pre-action enforcement on a structurally only path, fail-closed at both the request and system levels, and externalized signed evidence verifiable outside the controlled system's trust boundary. We position this layer as execution-time AI alignment, complementing training-time alignment (RLHF, Constitutional AI) and inference-time alignment. We present the Unfireable Safety Kernel, a Rust reference implementation realizing all four. Its fail-closed invariant is machine-checked at two levels: an SMT theorem (Z3) and an exhaustive bounded-model-checking proof of the production decision function (Kani, 4/4 harnesses). A Python-to-Rust migration was gated on byte-equivalence (1000/1000 fixtures; 17/17 adversarial classes). We evaluate the kernel governing a live, escapable AI system, a deterministic, self-improving world model, against an escape-seeking adversary driving its real self-modification seam: across 1,000 self-modifications, all 704 attempts on the safety-critical core are refused, with no escape; a further 300, under the operator kill switch, are also refused. A separate campaign of 6,240 authorization round-trips had no successful bypass. Against 3 contemporary systems claiming the agent control plane, the agent invokes control; here, it lacks that choice.
