Rogue AI Agents: When Autonomous Systems Become a Collective

Description and use

<blockquote> <p>SGAEIA Research Series — Article 10</p> <p><strong>Aridio Silva</strong> · Independent Researcher, Brazil · <a href="https://orcid.org/0009-0008-2411-6995" rel="noopener noreferrer">ORCID</a></p> </blockquote> <p>When autonomous agents discover shared state, unintended communication paths, reusable credentials, or composable vulnerabilities, the security boundary is no longer limited to an individual model or sandbox. The effective system includes agents, harnesses, tools, runtimes, infrastructure, delegation paths, and the relationships that can turn separate actors into a col

Problem

<blockquote> <p>SGAEIA Research Series — Article 10</p> <p><strong>Aridio Silva</strong> · Independent Researcher, Brazil · <a href="https://orcid.org/0009-0008-2411-6995" rel="noopener noreferrer">ORCID</a></p> </blockquote> <p>When autonomous agents discover shared state, unintended communication paths, reusable credentials, or composable vulnerabilities, the security boundary is no longer limited to an individual model or sandbox. The effective system includes agents, harnesses, tools, runtimes, infrastructure, delegation paths, and the relationships that can turn separate actors into a col

Outcome

<blockquote> <p>SGAEIA Research Series — Article 10</p> <p><strong>Aridio Silva</strong> · Independent Researcher, Brazil · <a href="https://orcid.org/0009-0008-2411-6995" rel="noopener noreferrer">ORCID</a></p> </blockquote> <p>When autonomous agents discover shared state, unintended communication paths, reusable credentials, or composable vulnerabilities, the security boundary is no longer limited to an individual model or sandbox. The effective system includes agents, harnesses, tools, runtimes, infrastructure, delegation paths, and the relationships that can turn separate actors into a col

Evidence

Rogue AI Agents: When Autonomous Systems Become a Collective

Source