Your AI agent will patch itself. The question is who approves the patch.

Description and use

<p>In the first week of September three arXiv papers landed on the same nerve:</p> <ul> <li> <strong>HarnessDev</strong> (2609.01437) asked whether LLMs can create and evolve their own agent harness.</li> <li> <strong>SafeEvolve</strong> (2609.02786) had agents co-evolve their own safety policy from experience.</li> <li> <strong>PatchBench</strong> (2609.04075) measured how well agents patch real security vulnerabilities — and found the usual proof-of-concept check inflates solve rates by 1.83×.</li> </ul> <p>Read together, they say one thing: agents are starting to modify the thing that runs

Problem

<p>In the first week of September three arXiv papers landed on the same nerve:</p> <ul> <li> <strong>HarnessDev</strong> (2609.01437) asked whether LLMs can create and evolve their own agent harness.</li> <li> <strong>SafeEvolve</strong> (2609.02786) had agents co-evolve their own safety policy from experience.</li> <li> <strong>PatchBench</strong> (2609.04075) measured how well agents patch real security vulnerabilities — and found the usual proof-of-concept check inflates solve rates by 1.83×.</li> </ul> <p>Read together, they say one thing: agents are starting to modify the thing that runs

Outcome

<p>In the first week of September three arXiv papers landed on the same nerve:</p> <ul> <li> <strong>HarnessDev</strong> (2609.01437) asked whether LLMs can create and evolve their own agent harness.</li> <li> <strong>SafeEvolve</strong> (2609.02786) had agents co-evolve their own safety policy from experience.</li> <li> <strong>PatchBench</strong> (2609.04075) measured how well agents patch real security vulnerabilities — and found the usual proof-of-concept check inflates solve rates by 1.83×.</li> </ul> <p>Read together, they say one thing: agents are starting to modify the thing that runs

Evidence

Your AI agent will patch itself. The question is who approves the patch.

Source