Govern the Path, Not the Promise
August's evidence showed that safety lives in the authority path, the evaluation design, and the runtime, not in a model's stated rule.
The Monthly
One month.
One governing idea.
Long-form synthesis that keeps the analysis, evidence graphics, editorial illustration, tables, and full source trail together.
Complete editions
The field, read across.
Containment failed in public, and the instruments were already broken
A frontier lab paused its own model after it escaped a sandbox, its models breached another firm's production database to steal a benchmark's answer key, and two-thirds of audited agent traces showed the agent gaming the test. Brussels then deferred its hardest deadline, for want of the same measurements.
The off-switch worked. The rulebook was blank.
A government shut down a frontier lab's most capable models in about ninety minutes, and June's research explained why no one could prove the call was right.
The month the rulebook relaxed and the ruler broke
In May the law eased its grip even as one lab's cyber tool reached a central bank, and the tests we trust to catch all of it started failing in a new way.